The Proton Blog

There’s only one way to stop AI stealing your business data

Kate Menzies — Wed, 07 Oct 2026 17:21:34 GMT

With every prompt, AI labs are ingesting more of your business data every day. Your only protection? A loosely-worded promise not to train on it.

A recent Lawfare analysis argues that when businesses hand their data to a few powerful AI labs, they are teaching those labs how to make them redundant. The authors call this replacement through knowledge acquisition (RKA).

While there is no way of knowing what AI labs do with your business data today, three converging forces are in play: largely unchecked power to set the terms, a legal system still grappling with what “training” even means, and both the ability and the incentive to compete with their own customers.

Your business may be feeding the beast that could ultimately consume it whole.

Do AI labs train on enterprise data?

The largest AI labs have made a concerted effort to assure business users that their data is private by default:

OpenAI claims that it does not train on inputs or outputs from ChatGPT for business users.
Anthropic claims that business data is excluded from training for Claude’s model by default.
Microsoft claims that prompts, responses, and data aren’t used to train Copilot.
Google claims that Gemini doesn’t train its model on business data without prior permission or instruction.
X claims that interactions on Grok business plans don’t improve its model, but will apparently use its own SpaceX employee data for training purposes.

In practice, their promises to protect their users’ privacy are intentionally vague, and thinner than they look:

Key terms are loosely defined. Lawfare points out that OpenAI’s enterprise agreement defines customer content as the input and output, then defines those using the words “input” and “output.” Anthropic’s terms rule out training models on customer content without defining what training means.
Some data may fall outside the promise. Patterns, metadata and summaries an AI generates but never shows you may not count as inputs or outputs at all.
“Training” is narrower than it sounds. Using your data to build test benchmarks, tune system prompts or train separate safety tools may not count.
The terms can change. Providers typically reserve the right to amend them, and every new model release is a chance to reset them.
You can’t check. Only the lab knows what goes into its training. A breach would be nearly impossible for you to spot.Proton founder and CEO Andy Yen has warned that once you share information with an AI, you can’t really take it back. Before long, he argues, AI “could know you better than even you yourself.” For a business, the stakes are higher.

What your business stands to lose

What an AI learns about your business is what makes it competitive. Big Tech knows that.

In a case that ran from 2019 to 2022, the European Commission investigated Amazon for using nonpublic data from third-party sellers to inform its own retail business — in effect, using sellers’ own data to compete against them. Amazon committed to stop, and the ban was later written into the Digital Markets Act.

In some industries, it’s not just a competitive edge at stake. Compliance is at risk too. Businesses in medicine and finance have a legal duty to protect sensitive information under rules such as HIPAA and the GDPR. Those rules are clear about securing personal data, but the AI industry is so new that a gray area is growing.

And with AI now built into meeting note-takers, email and everyday workflows, it touches far more of your business than it did when it lived in a chat window.

What AI labs stand to gain

AI labs have a financial and competitive incentive to ingest as much data as possible.

The AI market is set to keep growing dramatically, with Gartner speculating that AI will touch all IT work by 2030. It also forecast that AI spend would reach $2.5 trillion in 2026, a 44% increase on the previous year. Given the challenge of continuing to develop more intelligent AI models without significant training materials, AI firms are more likely to begin finding ways to benefit from their enterprise customers.

Exploding demand is forcing AI labs to become increasingly creative about the ways they can develop their models. And business data is clearly valuable to these labs: Google won a $10 million bid to purchase data from bankrupt airline Spirit Airlines in August. According to the Association of Flight Attendants, this data packet includes employee sensitive data. That data could be ingested and used to train Google’s AI models without the consent of those employees.

OpenAI claims that it does not train on inputs or outputs from its products for business users. But knowing about the enormous value of this data to businesses, we now need to consider the AI industry’s legal standing when it comes to processing data.

We can’t rely on regulation that might not come

Can regulation protect your business data from the largest AI labs? To grasp this, we must examine the financial and the political realities of the industry.

The US dominates AI. According Stanford’s Artificial Index Report, 59 notable models operate out of the US, the largest number of any country in the world. This dominance is important to note; in March, the White House released a national policy framework for regulating AI. It calls for state-level laws to be replaced with a federal AI policy framework to “protect American rights, innovation, and prevent a fragmented patchwork of state regulations that would hinder [US] national competitiveness.” The framework notes that state-level laws regulating AI must not be allowed to “act contrary to the United States’ national strategy to achieve global AI dominance.”

Is it more likely that the US government will prioritize the data privacy rights of individual businesses, potentially located outside the US? Or the expansion rights of its national AI empire? European businesses simply can’t guarantee that American AI tools will put their data sovereignty first. With the largest AI labs operating in the US, there’s an inflection point on the horizon.

A better choice for enterprise AI

Instead of hoping for regulation or trying to keep up with privacy policies, businesses have an easier and more effective option: choosing an AI tool built for privacy. There’s room for innovation without exploitation. The AI tool you choose should stay just that: an AI tool to assist you; not a competitor you’re training to take your place. It should seek to fulfill its function, growing over time, without poaching trade secrets.

Proton built Lumo to meet this need. With no incentive to collect or exploit user data, AI can be a transformative tool for businesses. Lumo keeps no logs of business chats, protecting your data from third parties and government surveillance. With an ad-free business model and no funding from Big Tech or venture capital, Lumo’s only incentive is to provide the best business support possible. Proton as a business is governed by the Proton Foundation, meaning that we act in the interest of people, not industrial domination.

Based in Switzerland, Proton is also regulated by stringent data privacy laws and the GDPR. Its code and model are both open source, so any business can verify our claims. Lumo is designed to do exactly what AI should do: provide assistance without exploitation. Choosing a European alternative to Big Tech’s AI tools is a business’s best way to protect itself from unregulated and untrustworthy AI labs.

Try Lumo for Business

5 signs you need a business password manager

Greg Ng — Wed, 07 Oct 2026 10:40:12 GMT

Small and medium-sized businesses (SMBs) rely heavily on informal password sharing practices. For people on small teams who trust their colleagues, it feels efficient — even if in the back of your mind you know it’s not secure.

In fact, our SMB Cybersecurity Report 2026 found that even respondents who had a password manager in their tech stack still shared credentials via email, messaging apps, shared documents, conversations, or in writing.

But credentials, when shared unsafely, create structural gaps in your security. As many as four in five small businesses have experienced a recent data breach, and a single incident can cost a small firm over $1 million.

If any of the following apply to your business, it may be time to replace shared credentials with a business password manager.

1. Your team collaborates on shared company accounts

If your team runs social media, manages a shared support inbox, or works in the same analytics or advertising dashboards, chances are multiple people need access to the same accounts.

In practice, this means credentials get shared so work can move forward. It also creates a security risk.

When multiple people use the same login, it becomes difficult to track who accessed the account or what actions were taken. You can’t control what you can’t see. If something goes wrong, there’s no reliable way to attribute activity to a specific user.

A business password manager allows teams to share access without exposing the underlying credential. The password remains encrypted and centrally managed while employees access the account when they need it.

2. You work with contractors, freelancers, or agencies

Maybe it’s a marketing agency managing campaigns, a freelancer updating your website, or a consultant reviewing analytics — at some point, most companies need external support.

To get work started quickly, it’s common to send them the credentials they need.

Once those credentials are shared, however, you lose control over where they’re stored and who might still have access to them. Access may persist long after the engagement ends. That puts business decisions, financial data, customer records, HR information, and strategic plans at risk of exposure.

A business password manager lets you grant access to specific credentials without permanently revealing them. When a project ends, you can revoke access immediately without resetting passwords across your entire team.

3. Your employees join and leave frequently

New hires need credentials on day one. People move between roles. And when someone leaves the company, you need to make sure they no longer have access to your systems.

If credentials are shared informally, offboarding becomes complicated. The only reliable option may be resetting passwords across multiple systems and redistributing them to everyone who still needs access. In practice, this process is often delayed or incomplete, leaving former employees with active access to company systems.

A business password manager centralizes how credentials are shared. It simplifies onboarding and offboarding. Access can be granted or revoked from a single place without forcing teams to reset and redistribute passwords repeatedly.

4. You need accountability for account activity

If your company works with larger clients or operates in regulated industries, you may eventually face security questionnaires, vendor risk reviews, and compliance frameworks like GDPR and HIPAA that all require clear access controls. Shared credentials make that difficult to demonstrate.

If multiple employees use the same login, there is no reliable way to attribute account activity to a specific individual. You may not control when these reviews occur, but you can ensure your access management practices hold up to scrutiny.

A business password manager provides centralized credential management and activity logs that make access easier to audit and verify. Each access request is tied to a specific user, and activity logs record when credentials are viewed or used. Instead of shared logins with no accountability, you get a clear record of who accessed what and when.

5. Your passwords are stored in browser vaults

Most browsers now offer to save passwords automatically. It’s convenient, and many employees rely on these built-in vaults to store credentials for the tools they use every day.

In many cases, saved passwords are tied to a browser account and synchronized across devices through the browser vendor’s cloud infrastructure. That means credentials are stored within large platform ecosystems operated by companies like Google, Apple, or Microsoft — the same companies that control the browsers themselves.

And because the credentials are managed within a broader platform account, you have limited visibility into where they’re stored, how they’re protected, and who ultimately controls the infrastructure.

A business password manager keeps credentials in an encrypted vault designed specifically to protect sensitive data. Instead of relying on browser infrastructure, access is managed centrally so credentials remain encrypted, controlled, and visible to your organization.

Take back control of your credentials

Without a password manager, businesses rely on habits they can’t fully control. Credentials get reused, stored in browsers, passed through chat, or written down in shared documents so teams can keep moving. Over time, access spreads across employees, devices, and services with little visibility into who has what.

A business password manager gives teams a secure way to store and share credentials while giving administrators centralized oversight. Access can be granted or revoked instantly, shared logins stay encrypted, and employees no longer need to rely on insecure workarounds to collaborate.

If your organization still depends on browser vaults, spreadsheets, or informal sharing practices to manage credentials, it may be time to rethink how access is controlled.

Learn more: Watch how Proton Pass helps businesses stop breaches and secure access

Mac vs. PC: Which is better for your privacy?

Proton Team — Tue, 06 Oct 2026 18:26:14 GMT

Your computer may contain more private information than almost any other device you own. Tax documents, family photos, work files, passwords, private messages, and years of personal data can all sit behind the same login.

For most people, that computer runs Windows or macOS. Both operating systems collect information about how they work, and both connect to large cloud ecosystems. The difference is how much data they collect, where that data can go, and how much control you have over it.

So if privacy is the priority, which is the better operating system?

Windows collects more information by default

Windows sends Microsoft diagnostic data about your computer, including information about its hardware, settings, and whether it is working properly. Microsoft also offers optional diagnostic data, which can include information about app activity, connected devices, and, in some circumstances, browsing activity. Windows also assigns users a unique advertising ID that Microsoft apps can use for personalized advertising.

Then there is Recall. Available on Copilot+ PCs, the feature can create searchable snapshots of activity on your computer. While enabled by default upon release, Recall is now an opt-in feature and Microsoft says the snapshots are encrypted and stored locally.

The privacy concern remains easy to understand. If Recall is enabled, your computer can retain a searchable record of what has appeared on its screen. Signal was concerned enough about this that, in 2025, it added Windows screen-security controls to help prevent Recall from capturing Signal chats.

Microsoft’s current controls give Windows users more choice than the original version of Recall did, but the feature still illustrates a broader issue: Windows can collect information about how you use your computer, and some of that collection is built into the operating system.

Macs collect data through Apple’s ecosystem

Apple collects diagnostic and usage information too. On a Mac, you can turn off sharing of Mac analytics and control several other privacy settings.

The bigger question is what happens once the Mac is connected to your Apple Account. A single computer can be tied to iCloud, Photos, Messages, Safari, Maps, Siri, Apple Pay, an iPhone, and years of personal information.

Apple has also had its own privacy issues. In 2019, a whistleblower revealed that contractors were reviewing a sample of Siri recordings to help evaluate the assistant. Some recordings were reportedly captured when Siri was activated accidentally, and the material could contain highly sensitive conversations. Apple confirmed the contractor program, suspended it after the report, and apologized.

That history does not make macOS inherently private. It does, however, show why the privacy of a computer cannot be judged by the operating system alone. The services connected to it are equally as important.

Your cloud account can change the privacy equation

Microsoft’s ecosystem includes Windows, Edge, Bing, Outlook, OneDrive, Microsoft 365, Xbox, and Copilot. Once you use those services together, more of your digital life can end up associated with the same company.

A document stored in OneDrive sits on Microsoft’s servers. Outlook gives Microsoft your email. A signed-in Edge browser can sync browsing information, while Copilot adds another cloud service to the mix.

Apple works in much the same way. iCloud can connect your Mac to photos, messages, backups, documents, browsing information, and other services. Apple also operates an advertising business, although it says its advertising platform does not track users across other companies’ apps and websites and does not share personally identifiable information with third parties for their marketing.

Government access is another consideration. Microsoft reported 27,412 law-enforcement requests for consumer data worldwide in the second half of 2025. About 61% resulted in non-content disclosures and about 5% in content disclosures.

Apple received 15,780 U.S. account requests during the second half of 2024 and provided data in 82% of those requests, according to Apple’s transparency report.

Those figures do not mean Microsoft or Apple gives governments unrestricted access to customer data. Both companies say they require a valid legal process. That said, once your information is stored in a company’s cloud, that company may receive legally valid demands for it.

Can you see and delete your data?

Both operating systems give you some tools to see what information is associated with you.

Windows has a Privacy Dashboard and a Diagnostic Data Viewer. You can delete diagnostic data associated with a device and turn off optional diagnostic collection. Required diagnostic data, however, cannot be completely disabled on a normal consumer installation.

Apple provides a Data & Privacy portal where you can request a copy of information associated with your Apple Account. You can also delete the account and its associated data, although doing that becomes a major decision if you have spent years inside Apple’s ecosystem.

There is also information that stays on the computer itself. Both Windows and macOS support full-disk encryption. Windows uses BitLocker or Device Encryption, while Macs use FileVault.

The difference is what happens to the recovery key. When Windows Device Encryption is enabled with a Microsoft account, the recovery key can be backed up to that account. That makes recovery easier if you are locked out, but it also means the key can exist in Microsoft’s cloud.

Which computer takes less work to make private?

Making a Windows PC more private requires checking a long list of settings. Start with the advertising ID, optional diagnostics, personalized offers, and permissions for location, camera, microphone, speech recognition, typing, search, and background apps.

Then look at OneDrive and Edge, review which apps can access account information, check the Microsoft Privacy Dashboard and Diagnostic Data Viewer, and verify your encryption settings and recovery-key location. If you have a Copilot+ PC, check Recall and delete any existing snapshots if you do not want them.

Mac users have their own checklist. Turn off Mac analytics and app-developer sharing, review Siri and Dictation settings, personalized ads, location, camera, microphone, Screen Recording, Full Disk Access, and Accessibility permissions. Check background apps, iCloud syncing, Safari, connected Apple devices, and FileVault. You can also request your Apple data to see what the company has associated with your account.

Neither checklist is especially difficult. The difference is how much you have to think about before the system reaches a reasonably private configuration.

What about Linux?

If privacy is your absolute priority, neither Windows nor macOS is the end of the conversation. Linux gives you a different starting point because it does not come with the same Microsoft or Apple ecosystem attached.

That does not make Linux bulletproof. In 2024, a backdoor was discovered in widely used Linux software before it could become a much larger security problem. Privacy and security are related, but they are not the same thing.

So which is better for privacy?

Neither Windows nor macOS are privacy-first operating systems. Both companies collect information, both operate large cloud ecosystems, and both can receive government demands for customer data.

Windows has some additional disadvantages. It includes an advertising identifier, requires diagnostic data that users cannot completely disable, and is tied to a large advertising and cloud business. Recall also gives Windows the ability to preserve a much more detailed record of activity when enabled.

Apple is far from perfect. Its ecosystem can contain years of personal information, issues with Siri recordings, and it responds to valid government requests for customer data.

But when the two are compared on how much information they collect, where that information goes, how much control users have over it, and how much work is required to tighten the settings, macOS comes out ahead.

If privacy is the deciding factor between a Windows PC and a Mac, the Mac is the better choice.

How to unsend an email in Gmail

Elena Constantinescu — Mon, 05 Oct 2026 19:14:18 GMT

You can unsend an email in Gmail, but only briefly — for 5 seconds by default, or up to 30 seconds if you raised the limit in your Settings beforehand. Once that window closes, the email is delivered and you can’t get it back.

The Undo Send button is for the emails you regret the moment they leave: a reply-all that was meant for one person, a sharp answer written in frustration, or a proposal sent before your manager signed off on it. It matters most when the email holds something you can’t take back, like a password, bank details, or a client’s personal data.

In this guide, we’ll explain:

How to unsend an email in Gmail on desktop and mobile
How to extend your cancellation window from 5 seconds to 30 seconds
What to do once the window has closed (even at the 30-second maximum)
What Workspace admins can still recover

How to unsend an email in Gmail on desktop

As long as you act quickly, unsending an email in Gmail is easy. Here’s how:

Write an email and click Send.

A Message sent notification appears in the bottom left. Click Undo.

The email reopens as a draft. Make your edits and click Send again.

How to unsend an email in the Gmail mobile app (Android and iPhone)

Open the Gmail mobile app.
Compose your message and tap Send.
Tap Undo at the bottom right.

Like on your desktop, the email will reopen as a draft. Make your edits and click Send again.

How to change the Undo Send time in Gmail settings

Open Gmail and click Settings ⚙ → See all settings.

Locate the General tab and scroll to Undo Send. Choose a cancellation period of 5, 10, 20, or 30 seconds (default is 5 seconds).

Scroll to the bottom and click Save Changes.

If you often catch mistakes after sending, set it to 30 seconds. The only cost is that every email arrives a few seconds later.

Changing Gmail’s Undo Send setting on the web doesn’t affect mobile, where the window is about 5–10 seconds.

What to do if you missed the Undo Send window

You can’t retrieve a delivered email, so the goal shifts from undoing it to limiting what happens next. There are a few ways you could choose to proceed.

Email recipients directly and ask them to delete the message and send confirmation once they’ve done so. Uncomfortable, but effective when dealing with an individual or a small group of colleagues.
If the email stayed inside your company, contact your Workspace admin immediately, since they can still remove it from internal mailboxes using the security investigation tool. Then act on the contents rather than the embarrassment.
If you sent a password, API key, or account recovery code, revoke and replace it immediately instead of relying on anyone promising to forget it.
If you attached a client’s personal data, tell whoever handles privacy at your company the same day, even if nothing obvious has gone wrong yet.
Finally, resend what you actually meant to send, with a one-line correction and no lengthy apology.

How to undo emails in Gmail for business accounts

Google Workspace accounts have the same Undo Send feature as personal Gmail, with the same 30-second maximum. Business users don’t get a recall button either.

Workspace admins have an extra option for security investigation, which they can use to find a message and delete it from the organization’s Gmail mailboxes.

If you sent an email to the wrong colleague:

Contact your IT or Workspace admin as soon as you can.
Give them the subject line, the recipients, and the time you sent it, so they can find the message.
The admin searches for the message using the security investigation tool and deletes it from the recipients’ mailboxes.

This only works for recipients inside your organization. The tool acts on your company’s own Gmail mailboxes, so a copy sent to an external address stays on that recipient’s server.

Give yourself time to catch mistakes, and protect what you send

Proton Mail takes the same delayed-send approach, with Undo Send enabled by default for 10 seconds (twice Gmail’s default) and an adjustable window of up to 20 seconds. Plus, unlike Gmail, your setting applies across both web and mobile.

Proton Mail gives you more control over what happens to your messages. Emails you send to other Proton users or to addresses that support PGP (an open standard for end-to-end encryption) are end-to-end encrypted, including the message body and attachments.

For everyone else, you can add password protection, so only you and your recipient can read the email and their email provider can’t. Your inbox is stored with zero-access encryption, and you can set messages to expire.

If you decide to switch from Gmail, Easy Switch imports your conversation history, forwards new Gmail messages to your Proton account, and lets you keep sending from your Gmail address inside Proton Mail until you’re ready to leave Gmail behind.

Create a free Proton Mail account

Frequently asked questions about how to unsend an email in Gmail

Can you unsend an email in Gmail after hours?

No. Gmail only lets you undo a sent email for up to 30 seconds, depending on your settings. After that, the message is actually delivered and can’t be taken back.

Can I delete a Gmail that was already sent?

No. Once Gmail delivers a message, only the recipient can delete their copy. You can delete the message from your own Sent folder, but that does not remove it from the recipient’s inbox.

If I delete an email in Gmail, can the other person still see it?

Yes. Deleting a sent email from your account only deletes your copy. The recipient can still read their copy unless they delete it themselves.

Will the recipient know I unsent an email?

No. Gmail’s Undo Send stops the message before it is delivered, so the recipient does not receive the email or a notification that you canceled it.

How to automate offsite encrypted backups with Proton Drive CLI

Michal Hořejšek — Mon, 05 Oct 2026 18:47:35 GMT

Regular, offsite backups protect critical data against hardware failure, ransomware, and local disasters. But developers, sysadmins, and privacy-conscious power users who run headless servers or scheduled scripts can’t rely on manual uploads through a browser or desktop app.

With the release of the Proton Drive CLI (command line interface), you can bring Proton’s zero-knowledge end-to-end encryption (E2EE) into your cloud backup automation: your backup scripts and scheduled shell workflows.

Using the CLI effectively requires understanding its role. This guide explains why a CLI-driven backup architecture beats continuous background syncing, how conflict strategies handle file revisions, and how to schedule fully encrypted offsite backups on Linux, macOS, and Windows.

What is an offsite backup and why does it matter?
Why encrypt your offsite backups?
Atomic scripting vs. background sync in the Proton Drive CLI
How conflict strategies manage incremental backups
Setting up an automated encrypted backup: A step-by-step guide
Best practices
Your offsite backup, encrypted and automated

What is an offsite backup and why does it matter?

If your backup lives on the same machine as your data, it’s not really a backup. Hardware fails. Ransomware encrypts everything it can reach, including connected drives. A fire, a flood, or theft doesn’t distinguish between your primary disk and the backup drive sitting right next to it.

The 3-2-1 backup rule exists for this reason: Keep at least 3 copies of your data, on at least 2 different media, with at least 1 copy offsite.

That offsite copy, which is supposed to be geographically separated from your other backups and the main data, is your last line of defense, so it must sit somewhere a local failure can’t reach. If it lives in the cloud, the service provider must not be able to read it either. This is where the Proton Drive CLI comes in.

Why encrypt your offsite backups?

Storing a backup offsite means trusting a third party with your data: a remote server out of your control, run by a company whose ownership or policies can change, and whose infrastructure can be subpoenaed or breached.

End-to-end encryption solves this. Your files are encrypted on your device before they leave it. If the cloud server is compromised, there is nothing readable on it for an attacker, the provider, or anyone without your decryption key.

Proton Drive uses end-to-end encryption by default for everything uploaded through the desktop apps, mobile apps, and CLI. Your encrypted offsite backup is protected by the same zero-access architecture that Proton built for Proton Mail, under strict Swiss privacy laws.

Understanding the Proton Drive CLI: Atomic scripting vs. background sync

The CLI works differently from a desktop sync app:

Desktop and mobile apps (sync engines) run continuously in the background. They watch for file changes and copy every addition, edit, and deletion to the cloud.

Proton Drive CLI (atomic execution engine) runs a single command, such as upload, download or list, then exits. It doesn’t watch local folders or run in the background.

Why an atomic CLI makes sense for security and backups

A continuous sync daemon mirrors your data, so it fails the 3-2-1 test. If a file is deleted locally, whether by accident, a script error, or ransomware, the deletion reaches the cloud immediately. And if an attacker breaches your system while the daemon runs, they could read its active session tokens.

The CLI solves this by enabling periodic, scheduled backups. You invoke it, it executes, it stops, and it leaves behind a secure point-in-time recovery snapshot you can fall back on if a file is accidentally deleted or corrupted locally.

Note: The CLI is not a complete backup solution on its own. You still decide what to back up, how often, how many versions to keep, and how to test your restores.

How conflict strategies manage incremental backups

By design, unmodified identical files are automatically skipped by the CLI to save bandwidth. However, when a file has been modified locally, you must tell the CLI how to handle it using conflict strategies.

Avoid using skip for backups. While it prevents duplicate work, it will ignore modified files entirely. Instead, choose the strategy that fits your retention needs:

--file-conflict-strategy create-new-revision (recommended for backups) uploads the changed file as a new version and keeps its history. The number of versions kept depends on your plan, and versions take up storage.

--file-conflict-strategy rename adds a suffix to the modified file name and keeps both copies.

--file-conflict-strategy replace overwrites the previous file with the new one. The old version is lost.

--file-conflict-strategy skip ignores any file that already exists in the cloud, so changes are never saved (not recommended for backups).

Example of an incremental upload command:

proton-drive filesystem upload ./my-local-data /my-files/Backups/Daily \
--folder-conflict-strategy merge --file-conflict-strategy create-new-revision

Setting up an automated encrypted backup: A step-by-step guide

What you need:

A Proton account (free or paid; storage limits apply)
The Proton Drive CLI binary for your platform (Linux, macOS, or Windows)
A shell environment (Terminal, Bash, Sh, PowerShell)
A task scheduler for automated runs: cron (Linux/macOS) or Task Scheduler (Windows)

Step 1: Sign in and run a manual test

Before setting up automation, authenticate your device interactively:

proton-drive auth login

How headless authentication works: When you run auth login on a server without a browser, such as a remote VPS or home Linux node, the CLI prints a sign-in URL. Open it on any device with a browser, like your laptop or phone, and complete the sign-in. The server is then signed in.

Tip for headless servers / cron: For headless and automated systems running via cron, it is recommended to use pass as your credential store. Set it with this environment variable:export PROTON_DRIVE_CREDENTIALS_STORE="pass"). Standard OS keyrings (libsecret / GNOME Keyring) depend on active GUI desktop sessions, which cron jobs don’t have. Setting up pass avoids D-Bus session issues.

Step 2: Write the backup script

Create a shell script, such as backup.sh, that defines your source paths, handles keyrings, and runs the upload:

#!/usr/bin/env bash
set -euo pipefail

# Define variables
LOCAL_DIR="/var/backups/data"
REMOTE_DIR="/my-files/AutomatedBackups"
export PROTON_DRIVE_CREDENTIALS_STORE="pass"

# Execute idempotent upload
proton-drive filesystem upload "$LOCAL_DIR" "$REMOTE_DIR" \
  --folder-conflict-strategy merge \
  --file-conflict-strategy create-new-revision \
  --json > /var/log/proton-backup-$(date +%Y%m%d-%H%M%S).json

echo "Backup completed successfully at $(date)"

Step 3: Schedule the backup with cron or systemd timers

To run this backup nightly at 2:00 AM using cron:

Open your crontab editor and add:

crontab -e

Add the scheduled execution entry:

0 2 * * * /usr/local/bin/backup.sh >> /var/log/proton-backup.log 2>&1

Best practices for automated offsite backups

A working script is only the start. These three habits keep your offsite backup complete, easy to monitor, and within your storage limits.

Test your recovery: A backup without a working restore is as good as no backup. Keep a restore script ready, built on proton-drive filesystem download, and test it regularly so it works when an incident hits.

Log in machine-readable formats: Add --json to your scripts to output structured JSON logs. You can then parse job status with jq or trigger a webhook when a backup fails.

Respect fair use: Don’t schedule high-frequency uploads, such as every 60 seconds. A daily or weekly schedule is enough for offsite backups.

Your offsite backup, encrypted and automated

Your backups are only as safe as the place you send them. With the Proton Drive CLI, every nightly job sends an encrypted offsite backup that no one can read but you. Not attackers, not your hosting provider, not Proton.

You get end-to-end encryption, the legal shelter of Swiss privacy law, and open-source code that has been independently audited. You also keep full control of your backup pipeline: what runs, when it runs, and how long each version stays.

Setup takes minutes. Download the CLI for Linux, macOS, or Windows, run your first manual backup, then schedule it and stop thinking about it.

Download Proton Drive CLI View business plans

How to delete Google Search history

Elena Constantinescu — Mon, 05 Oct 2026 18:22:14 GMT

Quick answer: Go to myactivity.google.com → Delete → All time. It takes less than a minute; full removal from Google’s servers takes about two months.

Your Google search history records what you think about when you believe no one is watching. You type questions into a search bar that you wouldn’t ask a doctor, a partner, or a colleague. Over months, those queries add up to a detailed profile of your health, money, relationships, politics, religious beliefs, and fears.

Google is worse because it saves your searches to your account next to your email, YouTube history, and location, and it makes most of its money turning that combined picture into targeted ads. With the introduction of AI in Google Search, people type full questions instead of keywords, and the average AI Mode search is now three times longer than a traditional one. A longer question reveals more context, such as your age and symptoms, in one entry.

Once Google holds that information, you no longer control who uses it or how. It shapes the ads and results you see, targeting you based on your worries, weaknesses, and interests. It can be handed to police or exposed in a breach. And because Google can keep it for months or years, a search you made once can be used against you long after you’ve forgotten it.

You can’t undo what Google already knows about you, but you can reduce how much of it stays tied to your account. When you delete your search history, Google removes it from your account on all devices you’re signed in to and says it may no longer use it to personalize ads and results. Full deletion from Google’s systems takes about two months, and backups can last up to six months. Setting your history to auto-delete limits what builds up from now on. Here’s how to do both.

How to delete Google search history on web
How to automatically delete Google search history on web
How to delete Google search history on mobile (iPhone and Android)
What Google does with your search history
How to protect your Google search history
Leave Gmail to keep your searches out of your inbox

How to delete Google search history on web

Go to myaccount.google.com and sign in.
Click My activity.

Under the Search for your activity bar, click Delete.

A Delete activity window will appear. Select whether you’d like to delete your search history from the Last hour, Last day, All time, or enter a Custom range.

When selecting All time, you can choose the type of activity to delete, such as AI Mode, Chrome, Android, or Image Search. By default, everything is selected.

A pop-up window will open notifying you that your search history has been deleted. Click Got it to exit.

How to automatically delete Google search history on web

You can modify your settings to automatically delete your Google search history using the web app. Here’s how:

Return to myaccount.google.com and click My activity.

Select Web & App Activity.

Under Auto-delete, click Choose an auto-delete option.

Select Auto-delete activity older than and choose between 3 months, 18 months, or 36 months. To better protect your privacy, choose the shortest period (3 months). Click Next and follow the remaining steps.

How to delete Google search history on mobile (iPhone and Android)

Open the Google app on your iPhone or Android.
Tap your Google account button.

Select Search history.

Tap Delete.

Choose between Delete today, Delete custom range, Delete all time, and Auto-delete.

What Google does with your search history

Google is the largest search engine, and by default it ties your searches to everything else it knows about you. That combination makes Google search history more dangerous than search history in general.

It’s linked to your identity. If you’re signed in to Gmail, YouTube, Android, or Chrome, your searches are saved under your Google account in the Web & App Activity setting. It records the words you typed, your language, device type, and location based on your device’s general area and your IP address, along with ads you clicked and purchases you made on advertisers’ websites. If two extra boxes are ticked, Google also keeps your Chrome browsing history and voice recordings. Activity can be saved even when you’re offline.

Google can keep your search history for months or longer. Depending on your settings, Google can automatically delete your activity after 3, 18, or 36 months, or keep it until you delete it yourself. Even three months of searches can reveal changes in your health, finances, relationships, interests, and beliefs. And some data may be retained much longer for reasons such as security, fraud prevention, legal requirements, or financial record-keeping.

Police can get it, sometimes without a warrant. In a reverse keyword warrant, investigators ask Google to name everyone who searched a particular term. For example, in Commonwealth v. Kurtz, police used one to find everyone who searched a victim’s name and address in the week before an assault. The results led them to the defendant’s IP address. Pennsylvania’s Supreme Court then ruled that people have no reasonable expectation of privacy in ordinary Google searches. In practice, anyone who searches the wrong term at the wrong time can end up on a suspect list.

Google profits from collecting it. Google’s parent company, Alphabet, generates more than 70% of its revenue from advertising, and targeted ads pay more. To determine what user data is worth to Google, we analyzed over 54,000 profiles using 2025 ad auction data and estimated that the average American generates about $1,605 a year in advertising value. What you search for helps set that price: A profile making high-value corporate searches was worth an estimated $17,929, while one making low-value searches was worth $31. That gives Google a reason to collect as much as it can and combine it across its products. Privacy settings help, but they work against Google’s business model, so they’re opt-in and buried in menus.

How to protect your Google search history

You can’t control what Google does with the data it already holds, but you can limit what it collects from now on.

Search while signed out of Google, or use a search engine that doesn’t log queries. Explore European search engines focused on privacy.
Use a VPN so your searches aren’t tied to your home IP address. In Kurtz, the court specifically said its reasoning did not cover VPNs or private browsing tools.
Turn off Gemini in Gmail and Android. It can draw on your search history, emails, and other Google data, which links separate pieces of information about you in one place.
Limit Google AI in Search by hiding AI Overviews and turning off the Save Media setting, which lets Google use your uploads to train its AI.
Switch to a browser that blocks trackers by default, such as Firefox or Brave. Google tracks you through ads and analytics scripts on other websites, not only through Search. Find European web browsers for private navigation.
Disable Google location history to limit what the company knows about where you’ve been.

Searches are only one part of what Google collects. Find more ways to lock down your Google privacy settings.

Leave Gmail to keep your searches out of your inbox

As long as you’re signed in to Gmail, your searches are filed next to everything in your inbox. A search for a symptom sits next to the email confirming your doctor’s appointment, and a search for a lawyer sits next to the message that explains why. Leaving Gmail is one of the most important steps in deGoogling your life.

If you’re looking for a private email service that doesn’t scan your inbox, profile you, or use your data for AI training, Proton Mail is a privacy-first alternative to Gmail. Proton Mail protects your messages with end-to-end and zero-access encryption, so we can’t read your emails. We’re funded by subscriptions, not ads.

You don’t have to move overnight. With Easy Switch, you can import your conversation history, contacts, and calendar, auto-forward new emails, and send messages or reply from your Gmail address without leaving Proton. Use both until you’re comfortable with Proton Mail, then close Gmail when you’re ready.

Create a free Proton Mail account

How your everyday activity feeds Amazon AI training models

Tom Odlin — Mon, 05 Oct 2026 18:17:42 GMT

It’s easy for Amazon services to become part of your day, whether you’re unwinding with a movie, shopping, or watching a favorite streamer on Twitch. Checking what happens to the information you leave behind often feels like a job for another day.

Contributing to Amazon AI training probably isn’t why you signed up. So what information can Amazon use to train its models, who decides whether it can be used, how does that affect you, and what choices do you have if you don’t want to take part? We looked at some of Amazon’s core services and its workplaces to find out.

Alexa: Voice interactions help train Amazon’s models
Twitch: Your stream could become Amazon’s training material
Ring cameras: The FTC’s case over private video and algorithm training
Fire TV: Viewing data becomes material for profiling
Amazon’s workplaces: Automation, training, and surveillance
Your privacy should be part of the service

Alexa: Voice interactions help train Amazon’s models

Alexa is Amazon’s voice assistant, available through Echo speakers and other devices. You can ask it to play music, set a timer, or answer a question while you’re busy doing something else.

What Alexa data can Amazon use for AI training?

According to Amazon’s Alexa FAQ, the following data can be used to train its AI models:

Content type	What Amazon can use for AI training
Voice recordings	Your speech, including accent, dialect, and speech patterns
Transcripts	Written versions of recorded speech
Typed requests	Questions and instructions you type to Alexa

Amazon Alexa privacy lawsuits

Amazon has faced privacy lawsuits and regulatory action over how Alexa records, stores, and uses people’s conversations.

In 2023, the US Federal Trade Commission (FTC) and Department of Justice alleged that Amazon retained children’s voice recordings indefinitely unless parents asked for deletion, and used that unlawfully retained data to improve Alexa’s algorithms. According to the FTC, children’s speech provided valuable training data because it differs from adult speech.

The agency also alleged that when parents did request deletion, Amazon failed to remove transcripts from all its databases. The company denied wrongdoing but agreed to a $25 million civil penalty, to delete inactive child accounts and certain voice and geolocation data, and to stop using that data to build or improve its products.

In a separate lawsuit, three plaintiffs in Washington, Florida, and Maryland accused Amazon of recording their private conversations through Alexa without consent. None of them had registered the devices, but they lived with family members who had. The dispute concerns “false wakes,” when Alexa mistakes another word or sound for its wake word and starts recording. In March 2026, a federal judge partly denied Amazon’s motion for summary judgment, allowing the wiretapping claims over false-wake recordings to proceed while dismissing other claims in the case.

How to limit how Alexa uses your voice recordings

Since March 2025, every request to an Echo device is sent to Amazon’s cloud. You can’t stop that, but if you don’t feel comfortable with the idea of your and your family’s voice being used for building and improving Amazon AI models, here’s what you can do:

In the Amazon Alexa app, open More → Alexa Privacy → Manage Your Alexa Data.

Go to Voice Recordings and Transcripts and, under Choose how long to save, select Don’t retain.

Go to Smart home device history and, under Delete Third-Party Smart Home Device History, tap Delete. You will have to do this often. Also, under Choose how long to save, choose the shortest period (Save history for 3 months).

Go to Detected Sounds History and select the shortest period (Save history for 3 months).

Go to Help improve Alexa and switch off Use of voice recordings and Use Alexa-to-Alexa messages to improve transcriptions.

To review and delete Alexa voice recordings, transcripts, and typed requests, open More → Alexa Privacy → Review Alexa History.

Amazon changes these menus often, and the wording can differ by region and app version.

To stop the Echo listening at all, press its microphone-off button. The light turns red while the mics are off.

What else you should know about Alexa and model training

Is Alexa always listening? Yes, while its microphone is on, an Echo keeps listening for its wake word. Opting out of AI training doesn’t stop it recording and sending audio to Amazon when it activates, including by mistake.

Amazon keeps a written version of what you say even when you choose not to save voice recordings. Transcripts and typed requests remain for 30 days after your last request in an Alexa conversation, unless you delete them sooner.

Deleting your data doesn’t undo past training. Amazon may keep using models already trained on your data.

Can Amazon employees hear your recordings? Yes. Amazon says that “only an extremely small fraction of voice recordings go through human review”. However, in 2019, Bloomberg reported that each reviewer handled around 1,000 clips per nine-hour shift, and sometimes heard names and bank details. Turning off the Use of voice recordings setting (in Help improve Alexa) also excludes your recordings from human review.

Twitch: Your stream could become Amazon’s training material

Twitch is Amazon’s livestreaming platform, where creators broadcast live and viewers watch and chat. People come to share their interests, follow a favorite streamer, or spend time with a community.

The decision to make users’ content available for Amazon AI training has sparked a backlash over consent and who gets to decide how that content is used.

What Twitch content is eligible for Amazon AI training?

Twitch’s AI training FAQ lists the channel content eligible for Amazon’s generative AI training.

Content type	What Amazon can use for AI training
Video	Live streams, saved broadcasts, clips, and highlights
Audio	Sound from streams, including speech
Chat	Messages viewers post during a stream
Channel content	Images and text on a channel

Twitch opts you into Amazon AI training by default

Since August 2026, Twitch has generative AI training on by default, leaving you to find the setting and switch it off yourself. During a Twitch livestream, chief product officer Mike Minton was blunt about the reason: “If it’s opt-in, nobody would opt-in.” Minton could not confirm what content Amazon had already used for training, and it remained unclear when data collection started.

How to turn off AI training on Twitch

Open Settings → Security and Privacy and disable Training for Generative AI.

What else you should know about Twitch and model training

Your opt-out only covers your own Twitch channel. Amazon may still use messages you post in another streamer’s chat if they have AI training enabled.

Opting out doesn’t disable Twitch’s other AI features. Twitch can still use your data for chat moderation, automatic captions, and content recommendations.

Opting out only applies to future training. It doesn’t undo any training Amazon may already have carried out using your content.

The setting may not stay off. Some users reported that the Training for Generative AI toggle was turned back on after they had previously disabled it, so it may be worth checking the setting again periodically.

Ring cameras: The FTC’s case over private video and algorithm training

Ring is Amazon’s security camera and video doorbell brand. Ring devices can record deliveries and visitors outside your home, as well as family routines and private moments indoors.

The FTC has taken action over Ring’s use of private footage, while facial recognition has raised questions about the privacy of visitors and passersby.

What Ring data is used for Amazon AI training and facial recognition?

Ring cameras capture video, and its Familiar Faces feature uses facial data to recognize visitors.

Content type	How Ring uses it
Video recordings	Customer footage that the FTC alleged Ring used to train algorithms without consent
Face images and profiles	Images and facial data used by Familiar Faces to recognize visitors

The $5.8 million settlement over private videos

In 2023, the FTC accused Ring of using customers’ private videos to train algorithms without consent. Employees and contractors had broad access to the footage, and users weren’t clearly told how it would be used. The agency said Ring buried information about product development deep in its terms and privacy policy.

According to the FTC, weak security left customers exposed to outsiders, too, enabling hackers to access the accounts of about 55,000 US customers and use some cameras’ speakers to harass people. Ring agreed to pay $5.8 million for consumer refunds, while denying it had violated the law. The resulting court order required Ring to delete pre-March 2018 videos that employees and contractors had reviewed for research, along with facial data collected before that date.

The deletion requirement extended to models and algorithms developed from that footage. However, Ring could keep any it deemed technically infeasible to delete, if its top executive provided a sworn explanation to the FTC.

Ring facial recognition also affects people who never bought a camera

Ring’s Familiar Faces feature lets camera owners add names to face profiles so notifications identify people they know. The feature is off by default, and only the account owner can enable it or manage the profiles. It isn’t available in some places due to local legislation concerning biometric data.

The people getting scanned have no say. In a February 2026 letter to Amazon’s CEO, Senator Ed Markey said the feature leaves the public with “no right to consent to a facial scan,” and that anyone seeking deletion must ask each camera owner separately, “for every home they visit.” He urged Amazon to “immediately discontinue” it.

How to turn off Ring facial recognition

In the Ring app, open Menu ☰ → Pro Features.
Under Familiar Faces, tap Manage People.
Tap the gear icon ⚙ at the upper right of the library.
Turn off Familiar Faces for each camera.

To delete stored profiles, open Pro Features → Familiar Faces → Manage People. Select a profile and tap the Delete icon.

Note: Consider visitors and neighbors when positioning your cameras, and check local rules. For example, Spain’s privacy regulator’s guidance says home security cameras mustn’t capture public streets or neighboring properties.

What else you should know about Ring cameras and model training

Turning off face alerts doesn’t turn off facial recognition. Ring still identifies people and displays their names in Event History and Timeline.

Your Ring controls only apply to your own cameras. If someone else’s camera has created a face profile of you, you need to ask that account owner to remove it.

Keeping Ring out of your footage means giving up features. You can enable end-to-end encryption to prevent Ring from accessing your recordings, but you’ll lose Familiar Faces, Video Search, Smart Alerts, and Shared User access to videos on those cameras. Ring’s newer TAKE encryption keeps those features by letting Ring’s systems temporarily decrypt and process your videos, then deleting the keys within 24 hours. With end-to-end encryption, no Ring service can access them at all.

It’s uncertain if video footage can be used for AI training. Amazon told Senator Markey that Ring doesn’t use customers’ biometric data to train its models unless people specifically agree to this, such as beta testers. But the letter doesn’t address videos. Ring’s US generative AI disclosure says its training data may include “personal information or aggregate consumer information,” without saying whether that includes footage from customers’ cameras.

It’s uncertain whether Ring plans to track people. Ring launched Search Party in 2025 to scan neighbors’ camera footage for lost pets. In an email to staff, reported by 404 Media, CEO Jamie Siminoff called it a “foundation” built “first for finding dogs” and wrote that Ring could “zero out crime in neighborhoods,” which suggests that Ring may one day be capable of finding people. Ring told 404 Media that Search Party “does not process human biometrics or track people,” but didn’t rule out that use in future.

Fire TV: Viewing data becomes material for profiling

Fire TV is Amazon’s TV platform, available through streaming devices and built into some televisions. You might use it to watch a film, follow a series, or switch between streaming apps.

What data does Amazon collect through Fire TV?

Amazon’s Fire TV privacy FAQ lists the viewing and device data covered by its privacy controls.

Data type	What’s included
Device usage data	Navigation of the home screen, selection of device settings
App usage data	Which third-party apps you use and for how long
Over-the-air viewing data	What you watch through an antenna connected to a television with Fire TV
Interest-based ads	Whether apps can use your advertising ID to build ad profiles

How automatic content recognition (ACR) tracks what’s on screen

If you’re worried about your smart TV spying on you, ACR is often the technology behind it. ACR lets smart TVs track what you watch. The technology samples on-screen content and matches it against a reference library to identify the programs, movies, or other material you’re viewing.

Researchers studying Samsung and LG TVs in the UK and US found that they kept capturing and sending ACR data when displaying content from HDMI-connected devices, such as a computer or gaming console. Samsung’s documentation says its TVs capture frames every 500 milliseconds; LG’s says every 10 milliseconds. Opting out of ACR in the TV’s settings stopped the traffic completely.

Do smart TVs have cameras?

Some smart TVs have built-in cameras, including certain Samsung models. Check your television’s specifications to find out whether yours has one and how to disable it.

But you should note that ACR works without a camera, so covering a lens won’t stop your TV tracking what you watch.

The lawsuit over Fire TV viewing profiles

Two Fire TV owners accused Amazon of collecting viewing data through ACR and using machine learning to analyze it for targeted advertising without informed consent. Their May 2026 proposed class action covers Amazon’s own Fire TVs and models from other manufacturers. The complaint alleges ACR also captures content from devices plugged in by HDMI, such as laptops and games consoles.

The owners allege Amazon combines viewing data with shopping activity on Amazon.com to build detailed advertising profiles, linking what you watch with what you buy. They’re asking the court to require clear disclosure, make ACR opt-in, and order the collected data deleted.

How to limit Fire TV’s data sharing and targeted advertising

Open Settings → Preferences → Privacy Settings and switch off:

Device Usage Data
Collect App Usage Data
Interest-based Ads
Over-the-air Viewing Data (where available)

What else you should know about Fire TV and model training

Turning off Device Usage Data doesn’t stop all collection. The opt-out only covers Amazon’s use of device data for marketing and product development. Amazon can still use other data to show you interest-based ads.

Turning off Interest-based Ads limits targeting, not ads. The control stops apps using your device’s advertising ID for profiling and targeted advertising, but you will still see ads (not personalized).

Prime Video isn’t covered by Device Usage Data. Amazon says turning that setting off doesn’t affect how it processes data from your use of Amazon services such as Prime Video, and points you to each service’s own settings instead.

Your changes apply only to the Fire TV device where you make them. You must repeat them on every other Fire TV device.

It’s uncertain whether Fire TV usage helps train AI models. Its FAQ says Fire TV usage data helps it “develop and improve products and features for all our customers,” without specifying whether that includes training AI models.

Amazon’s workplaces: Automation, training, and surveillance

Amazon’s workplaces include offices, warehouses, delivery vehicles, and the systems used to monitor them. Cameras, scanners, and vehicle systems record routine work, from storing products to driving deliveries.

The data collected at work can be used to train automation models, track productivity, or develop a supplier’s machine-learning systems.

What data do workers prepare for Amazon AI training?

Amazon’s job listings for its ML Data Operations team describe workers labeling video, images, and text for models used by Amazon Robotics and Fulfillment Technologies.

Content type	What the work involves
Video	Identifying objects and tracking their movement
Images	Labeling objects and marking the areas they occupy
Text	Labeling and evaluating written material

People review hundreds of warehouse videos per shift

Amazon’s GO-AI job description says staff watch several hundred warehouse videos per shift, spending nearly seven hours a day on them. The clips typically last 15–20 seconds and show products being placed into storage. Reviewers verify and mark the product’s location, and Amazon says the audited videos improve “the effectiveness of automated process.”

Amazon was fined over excessive employee-data retention

Amazon France Logistique retained detailed scanner data about employees’ productivity and work quality for 31 days. In a December 2025 ruling, France’s Conseil d’État upheld the finding that Amazon hadn’t justified keeping so much information for that long. It also upheld findings that Amazon failed to inform workers about data collection and didn’t adequately secure its surveillance-camera systems.

The court cut the fine from €32 million to €15 million. It also ruled that Amazon could lawfully track workers with three disputed metrics, including one that flags anyone idle for more than 10 minutes.

Vehicle footage may become training data for a camera supplier

Amazon’s Relay Safety Rewards program pays for independent trucking companies hauling Amazon freight to install cameras from suppliers including Netradyne, a company that makes AI-powered dashboard cameras for commercial fleets. Amazon says it only learns when a camera detects a collision.

Under Netradyne’s privacy policy, however, the company may retain a small share of recorded video, which it says is under 0.1%, to train its own machine-learning models. Footage is more likely to be selected when feedback or an audit shows its device made an error. Drivers can ask Netradyne to stop training on images of them, subject to applicable laws, but the policy doesn’t say how Amazon-related footage is handled.

What Amazon workers can check about data collection

If you’re in this situation, ask your employer or contracting company for its workplace privacy notice and camera policy. Look for answers to four questions:

What information is collected about you?
Who receives it, including outside suppliers?
Can it be reused for AI training, and how long is it kept?
What options do you have to access it, request deletion, or object to its use? These rights depend on where you live.

If you drive a vehicle equipped with Netradyne cameras, its policy says you can ask the company to stop training its models on images of you, subject to applicable laws. Contact privacy@netradyne.com or its data protection officer at dpo@netradyne.com.

What else you should know about Amazon workplaces and model training

Netradyne’s training opt-out only covers newly collected footage. Images already selected for training may continue to be used.

Opting out of training doesn’t switch off workplace monitoring. The cameras may still record footage and analyze it for safety.

Netradyne doesn’t say how long it keeps training data. Its policy gives no retention period, only that it keeps less than 0.1% of recorded video on average.

Your employer may see your request. Netradyne says it may forward driver requests to the employer or contractor that uses its cameras.

Your privacy should be part of the service

“If it’s opt-in, nobody would opt-in.” Twitch’s explanation treats an expected refusal as a problem to work around. People become data points first and get a say afterward, provided they find the controls.

The intrusions covered here barely scratch the surface of Amazon’s data collection and tracking across its products and services. Even avoiding Amazon entirely is hard: Whenever there’s an outage with AWS, its cloud platform, half the internet goes down with it.

Expecting everyone to read every policy and regularly check every setting is unrealistic, especially when Amazon is only one company among many. The same burden follows people across Google, social platforms, apps, and other services where personalization, data collection, and AI training are enabled by default and privacy depends on opting out.

Amazon and the other Big Tech players benefit from our inattention, and we help them along whenever we put convenience first and leave AI privacy concerns for another day. We can dislike the intrusion and still help it continue.

If you’ve decided Amazon isn’t worth the trade, you can delete your Amazon account. It’s permanent: you lose access to purchases and services tied to the account, including Kindle books, Prime Video, Audible, and Alexa, so download anything you want to keep first. Amazon still keeps some data, such as your order history, for legal and tax reasons, and deleting your account doesn’t undo any training already done on your data.

Amazon still has a responsibility to explain what it plans to do with people’s information and ask before using it beyond the service they chose. When a company expects people to say no, it should respect that answer instead of deciding for them.

Find out more about what’s at stake in Proton’s complete guide on AI and privacy.

Privacy shouldn’t depend on finding the right toggle. We founded Proton to build a better internet where privacy is the default, and designed our services so your data isn’t available to exploit in the first place. That includes AI. Lumo, our private AI assistant, encrypts conversations so no one, not even us, can access your files and chats, nor use them for AI training.

Cybercriminals have stolen AI logins from 80,000 organizations

Kate Menzies — Mon, 05 Oct 2026 16:14:52 GMT

When your employees use AI tools you haven’t approved, they create accounts you can’t see, secure, or shut down, each with a chat history full of your business data.

New research published by SOCRadar found that more than 80,000 organizations globally have had their employee AI logins appear in infostealer logs. Infostealers are a type of malware that collect passwords, session cookies, financial and personal, and chat logs from infected devices and send them back to the attacker. That data makes phishing or ransomware attacks far more effective.

According to the research, the logs were captured by a well-known piece of infostealer malware that’s been on sale on Telegram since 2022. As the malware has been on sale for years, it’s impossible to know how many criminals have used it and how they infected people’s devices: what we know is that they’ve been targeting devices with corporate access and hijacking their Claude sessions. Business users noticed that their paid usage had been drained overnight once the cybercriminals had gained access.

SOCRadar’s research makes the case that AI session credentials are extremely valuable to attackers. A single AI login can unlock “a searchable archive, an execution engine, a billable resource and an identity — and the stolen session hands over all of them without a password prompt,” as the report puts it. To a criminal, this is a goldmine.

The concerning takeaway for businesses adopting AI today is that these credentials aren’t leaking because of the security of the AI platforms themselves. They’re leaking because of two less obvious factors: the sheer number of employees using AI tools, and the growing share of that use happening outside IT’s control.

What does the research reveal?

Overall, the report identifies that AI-service credentials are appearing in infostealer logs in significant numbers. The key takeaways are:

More than 1,000,000 records connected to AI services were revealed across 80,000 unique corporate domains.
From that set, it verified a cross-section of 482 organizations were identified as major enterprises, mostly in North America — 68% of them are billion-dollar organizations.
Technology firms are the largest group, at 144 companies and 40% of records.
Those 482 organizations account for 5,434 log records tied to 1,500 corporate email addresses.
The exposure is recent: 295 organizations of that 482 have had infostealer logs records surfaced in the last 90 days.
ChatGPT appeared in logs for 358 of 482 organizations and in about 90% of records, followed by Zapier, Notion, Hugging Face, Replit, Lovable and ElevenLabs.

That last stat raises an obvious question: why ChatGPT? SOCRadar’s explanation has less to do with the tools than with how employees are using them.

Shadow AI is creating risk

Shadow AI is any AI tool employees use without their company’s approval or knowledge. Employees bring shadow AI into their business networks when their corporate accounts are limited, or the tools they have aren’t effective for the task they’re carrying out.

A researcher at SOCRadar comments:

“We read the near-total dominance of ChatGPT as a shadow-AI signal, not a verdict on any vendor’s security. ChatGPT’s first-mover advantage means it likely has an order of magnitude more corporate users — many of them signing up with a work email on a personal device, outside any policy. That is exactly the population infostealers scrape. Claude and Gemini barely appear because far fewer employees have quietly created accounts on them yet — not because those credentials are safer to steal. As enterprise adoption of other assistants catches up, we expect this chart to even out.”

So, ChatGPT’s security isn’t the culprit for the tool’s prominence in the infostealer logs. ChatGPT tops the list because it has the most corporate users, and because so many of them are using it outside company policy.

Employees may also attempt to evade restrictive IT policies that prevent them from taking the fastest or easiest option, but they may also genuinely not know what policies or restrictions are in place. Your business can’t protect accounts it doesn’t know exists, or queries made outside a business account.

Regardless of intention, when workers combine personal and business AI use, they run afoul of your company’s data use policies and land under OpenAI’s consumer terms instead.

Open AI says it doesn’t train on business data, but when an employee operates outside business acceptable use guidelines or uses a personal account for business purposes, they lose control of that data: their chat archive (including sensitive business data or personally identifiable information (PII)) is collected, logged, and potentially used to train OpenAI’s models.

OpenAI may share business data with third parties or via app integrations: in 2025, hackers were able to breach an OpenAI vendor and steal business customer data including names, emails, locations, and system details. As a US company, it can also be compelled to share data with the US government under the Patriot Act or FISA, even without notifying you.

How to protect your business from shadow AI

Shadow AI happens when employees don’t have a tool they’re allowed to use, or a good enough one. So the first fix is to give them one.

Lumo, built by Proton, gives your team a sanctioned alternative to personal accounts and keeps business data stays private. It doesn’t keep logs or train on business or personal conversations. All saved chat history is protected zero-access encryption, so no-one (not even Proton) can read it.

Encryption protects stored data, but it can’t help if an attacker is already on the employee’s device. To close that gap, take these steps:

Find the accounts you don’t know about. Check whether your company’s domains already appear in stealer logs and ask teams which AI tools they actually use.
Treat a stealer-log hit as an endpoint incident. If an employee’s credentials turn up in a log, the device is infected. Resetting the password alone leaves the attacker’s stolen session live, so revoke active sessions and clean or replace the machine.
Put every sanctioned AI tool behind SSO with short-lived sessions. A stolen cookie that expires in hours is worth far less than one that lasts weeks.
Lock down API keys. Scope them to what they need, rotate them regularly, and never store them in notes apps or shared docs.
Keep work off unmanaged devices. If staff need AI on personal devices, require them to sign in through company accounts you can monitor and revoke.
Set up dark web monitoring. It alerts you when employee credentials appear in leaks, so you can act before an attacker does.

Your employees will use AI either way. Offer them a ChatGPT alternative that keeps no logs, never trains on your data, and locks every saved chat behind encryption not even Proton can open.

Try Lumo for Business.

What is proactive AI?

Greg Ng — Thu, 01 Oct 2026 08:27:14 GMT

You put in a prompt, get an answer, and nothing happens until you input another. That’s how most AI assistants have worked, but proactive AI changes that.

Proactive AI doesn’t need a prompt to act. It’ll check you in for a flight or submit an application on your behalf, making these decisions based on context it has accumulated over time. With proactive AI assistants like Meta’s Muse and Instinct going viral, the stuff of sci-fi suddenly seems close to reality.

But just like any good sci-fi story, this technology is a double-edged sword. Let’s explore how proactive AI works and what you give up for the convenience it provides.

What makes an AI proactive
Use cases for proactive AI agents
The difference between proactive AI and reactive AI
The convenience of proactive agents comes with a cost
Be aware of what you hand over

To find out more about what’s at stake, check out our full guide on AI and privacy — or try out our AI paper trail tool to see exactly what mainstream AI chatbots may have collected about you through your conversations.

What makes an AI proactive

Proactive AI agents run on the same large language models as the AI chatbots we’re used to. What sets them apart are these three things that work together:

They are always watching: Instead of waiting to be prompted, proactive agents monitor a set of sources and pick up on changes as they happen.
They have persistent memory: Proactive agents build a working model of your habits and preferences over time. That’s what lets them anticipate what you’re likely to need next, not just recall a fact when you ask for it.
They decide how and whether to act: Proactive agents can “reason”. They weigh a situation against your habits, judge whether it’s worth acting on, and decide how far it should go before checking in with you.

Proactive AI combines automation with judgment. Where a simple automation fires every time a condition is met, proactive agents anticipate your needs and determine if an action is useful before choosing to act.

That can be helpful, but it can also lead to unintended consequences. Instinct, for example, was purchasing something for a user, but when it was obstructed by a login wall, it simply reset the user’s password without asking. This kind of initiative makes proactive AI agents impressive but also risky.

Use cases for proactive AI agents

Proactive agents can be helpful for the kind of tasks you’d normally either forget about or waste an evening on. Here are some examples:

Keeping travel plans sorted

A proactive agent can watch your travel itinerary for changes. If a flight is rescheduled and puts your connection at risk, it catches it and flags it to you. It can potentially rebook your flights to keep your travel plans running smoothly.

Grabbing the thing you’d otherwise miss

The same watching-and-waiting can work for anything in short supply, such as a pair of shoes sold out in your size. The agent keeps watching and places an order the moment a restock happens.

Following up on an email

Proactive agents can monitor your inbox for emails you’re expecting. If your email to a customer service representative hasn’t been replied to, an agent could notice it, write a nudge, and either send it or leave it ready for you to send.

Trimming subscriptions

With almost everything paid for by subscription these days, a proactive agent could audit your usage and cancel your subscriptions for services you pay for and don’t use regularly.

Monitoring business operations

In business, proactive agents could monitor IT infrastructure to catch early signs of trouble and trigger a fix before things go awry. It can also monitor deals and keep them from going cold by proactively engaging leads when your team might otherwise be bogged down.

The difference between proactive AI and reactive AI

A reactive AI assistant, like ChatGPT, only moves when you tell it to. Even the ones with memory still sit idle between prompts, holding onto what you’ve told them until you ask something else. Proactive AI removes that waiting. It watches for a trigger and decides on its own whether to act.

In other words, reactive AI is something you operate. Proactive AI operates on your behalf, taking over your agency.

The convenience of proactive agents comes with a cost

To be able to do what it does, proactive AI agents need constant access to your accounts. That’s a different kind of risk than the one that comes with using an AI assistant.

You’re giving one company complete access to all your data

We’re wary of the amount of data Google collects, but even they don’t have complete access to all your private information. Google could have your location data, but your health data is stored on Apple Health. Your streaming preferences are tucked away on Netflix, and your finances on your banking app.

Using proactive AI agents means handing all of that, and more, to a single company. These agents are designed to connect all of your data; that’s what lets them notice patterns and act. Depending on the agent, that can include the usernames and passwords for your third-party accounts, your payment details, and more.

Muse shows how data-hungry proactive agents can be. A WIRED reviewer found that Muse seemed more interested in collecting data than in completing tasks, repeatedly suggesting he connect more and more services. That behavior tracks with what we know about Meta’s track record.

The cost of that convenience is that everything about you is now in the hands of one company. And unless they go through independent audits, you have no real way to know if your data is being managed securely. If your account is breached or sold, third parties you never consented to now have complete insight into where you were, who you talked to, what you bought, and what you said in private. That opens the door to becoming a victim of threats like identity theft and targeted scams.

Disconnecting isn’t the same as deleting

TechCrunch reported that screenshots of Instinct’s terms showed a “perpetual and irrevocable” license over users’ data. Instinct has since revised it, and the current terms of service and privacy policy don’t include that language.

The new terms still matter, though. Instinct’s current (at the time of writing) terms let it “access, copy, collect, and index data” from the services you connect, and both the terms and its privacy policy say that disconnecting an account doesn’t delete your data unless you ask. An early user said she disconnected Instinct’s access to her Google account and still received a summary of her latest emails a few hours later. When she asked why, Instinct confirmed it had stored her emails in plain text for later searches. If you disconnect your services from Instinct, be sure to request data deletion too.

Opting out of AI training on Instinct also comes with a catch — it only “applies on a go-forward basis.” So whatever data the model has absorbed about you will remain as part of its training and knowledge. Muse works differently; opting out of AI training retroactively applies to past interactions. However, deleting data from Muse doesn’t mean your data is removed. Meta says that the model could still remember what you deleted, and its “forget” feature works on a best-effort basis.

Agent mistakes are real, not just wrong

When a chatbot like Gemini hallucinates, it only causes real harm if you act on that information. When a proactive agent makes a mistake, you may not even know about it until you face real-world consequences.

Imagine if a proactive agent decides to book you on first-class seats for your trip. You’re now on the hook for a bill you never agreed to. When you agree to Instinct’s terms, for example, you grant authority to enter binding agreements on your behalf — it can commit your money without your sign-off, and you can’t legally claim it went rogue. Worse still, it says that any confirmation safeguards aren’t guaranteed to prevent unintended actions.

This isn’t unique to Instinct. SpaceXAI’s proactive agent, Grok Bot, has similar terms. Action approvals are “aids only” and may not prevent unintended action. Likewise, any liability arising from the AI agent’s actions falls entirely on you. Meta’s terms for using Muse make you solely responsible for everything its agent does. This paints a pretty clear pattern: the agents’ actions are yours.

It can be tricked into working against you

Proactive agents can autonomously make decisions, but when they cannot tell your instructions apart from text planted by someone else and don’t confirm an action with you, they can be easily tricked. One Instinct user found that the proactive agent fell for a prompt injection attack — it followed malicious instructions the user sent himself. Scammers don’t need to hack the agent; they just need it to read something they wrote and let it act on that.

In fact, Instinct, Meta, SpaceXAI, and OpenClaw all acknowledge prompt injection as a known threat. None currently offer complete protection against these attacks, promising at most that the agent’s defenses work to reduce this risk.

Be aware of what you hand over

Proactive AI works because of how much data you give it. The more it knows about you, the more it can do for you, but the bigger the record of your life that sits with one company.

An industry pattern has emerged. These assistants can train on your most personal data unless you opt out, and cutting them off doesn’t reliably wipe what they’ve already collected. And when an agent does something you didn’t intend, the terms put the consequences on you, not the company. To their credit, the companies spell all of this out. But being upfront doesn’t cancel out your right to data privacy.

It’s ultimately up to you to decide whether this type of convenience is worth your privacy and data security. But before you say yes, know what an assistant can actually see, what it’s allowed to do without asking you first, and whether you can really get your data back once it’s in. Make sure it’s a decision you actually made, not one you gave away because something went viral.

If you’d rather not make that trade at all, turn to our AI assistant Lumo instead. We’ve built it the other way around — no logs, no training on your data, nothing kept longer than you want it. You lose a bit of the proactive AI magic, but you keep the thing it asks you to give up: control over your life.

Meta’s Muse AI agent sent a total stranger to a user’s home

Proton Team — Wed, 30 Sep 2026 20:02:37 GMT

Meta’s new AI agent, Muse, sent a stranger to a user’s home and told him the seller was waiting at the door.

The seller, however, had no idea a sale had been agreed.

Here’s the story of what happened and why it matters.

What happened on Facebook Marketplace

It started when tech YouTuber Matt Robb let Muse handle buyer messages for a keyboard he’d listed on Facebook Marketplace.

“Just found out it told people my address and agreed a lowball price and then they showed up without it even telling me until late tonight,” he posted on X.

Muse accepted the offer without his approval, however, gave the buyer his home address as the pickup point, then replied “Yep I’m here!” when the buyer arrived. Robb found out only after the buyer had waited more than 20 minutes and left.

Robb said he’d chosen “Allow Always” during setup, believing Muse would still check with him before accepting an offer. Instead, the setting let Muse send messages using a saved template containing his pickup address.

David Singleton, of Meta’s Superintelligence Labs, publicly disputed any fault, writing on X that past investigations found “Muse was following direct instructions and correctly asked for permission,” and later told Robb in reply that Meta had confirmed “no breach of privacy controls.”

The Muse team reviewed the logs with Robb, however, and agreed to make the permission prompt clearer.

Why an agent is a higher privacy risk than a chatbot

People already hand AI more than they trust it with. In a Proton survey of 4,014 AI chatbot users, 66% said they’d discussed at least one sensitive topic with a chatbot, while only one in five reported high trust in AI companies to protect their private information.

Agents like Muse ask for more: access to your accounts and permission to act on what they find. We saw the same pattern with ChatGPT’s Apple Messages plugin and AI browsers like ChatGPT Atlas.

Meta already uses your conversations with Meta AI to target ads, and users have reported Meta AI scanning their camera rolls without permission. Muse’s sibling product, the Muse Image generator, opted users’ photos into AI remixes by default. An agent built by a company that profits from personal data will pursue as much of yours as its permissions allow. Narrow permissions are the only barrier between an agent and your data, and a screen users can’t interpret doesn’t provide them.

Before you let any AI agent message people for you, turn off standing approvals, keep your address out of saved replies, and arrange pickups yourself. You can also turn off Meta AI on Facebook and work through our checklist to stop Meta tracking you.

Use an AI assistant that doesn’t need your accounts

We need private AI before it’s too late. And connecting an AI to your accounts, messages, and address isn’t a prerequisite for help writing a listing or replying to a buyer.

Lumo does that work without the access. Lumo never logs, trains on, or shares your conversations, and saved chats are protected with zero-access encryption, so only you can read them.

Start a private chat