privacy

25 AI Apps Changed Their Privacy Policy Last Month

An October 2026 audit found 25 of 58 AI privacy policies changed in a single month. Here's what shifted, which apps now train by default, and how to check.

In any given month, you might expect a handful of apps to update their privacy policies. An October 2026 audit by Getvoibe — which re-read 58 AI app privacy policies in a single month — found that 25 of them had changed between September 10 and October 13.

That’s nearly one change every day, across services ranging from AI voice keyboards to medical documentation tools to general-purpose assistants. Some changes were cosmetic. Others quietly switched training defaults from opt-in to opt-out, or added clauses permitting data use that wasn’t allowed before.

This is not a hypothetical risk. It is an ongoing, active shift in how AI companies treat the data you give them — and most users have no idea it’s happening.


Why AI Privacy Policies Are Changing So Fast

AI products face a structural tension. Training better models requires more data. More data means accessing what users have already stored or generated. And accessing stored personal data for a new purpose — training an AI — typically requires updating the privacy policy.

The result is a policy landscape that shifts constantly, driven by:

  • Competitive pressure: If a competitor’s model improves because it trains on user data by default, there’s commercial pressure to do the same.
  • Regulatory positioning: Privacy laws in the EU and US are still catching up to AI training. Companies are locking in data rights before stronger rules arrive.
  • Product bundling: AI features are being added to existing apps. Each addition creates an opportunity to expand what the app can do with your data.

The pace of change means that even if you carefully read an app’s policy when you signed up, that reading may be out of date.


What the October Audit Found

The Getvoibe audit flagged several examples worth understanding.

Wispr Flow is an AI voice dictation app that replaces your keyboard with speech-to-text. Its help center — updated in October 2026 — states that training is on by default for Free and Pro users. Its privacy policy, however, describes training as something users opt into. The audit flagged the contradiction directly: the help center and the privacy policy say opposite things.

Why does this matter? Wispr Flow works by capturing everything you dictate — messages, documents, emails, voice notes. If training is on by default, then every word you’ve spoken into your keyboard is potential training data. Users who believed they had opted out based on the privacy policy may not have.

Heidi Health, a medical AI documentation tool, began training on data from free-plan users. Medical notes are among the most sensitive data a person generates. A policy change that expands training to free-tier accounts, applied retroactively to existing users, is exactly the kind of change that’s easy to miss.

Apple added an explicit opt-in in the same period that allows it to train models on Siri requests. This is a more user-friendly change — it’s opt-in rather than opt-out, and Apple’s privacy posture for on-device processing has historically been more conservative than its cloud competitors.

But the Apple example illustrates the broader point: even companies with good track records are navigating this terrain, and the defaults are what matter for most users who never visit the settings page.


The Default Problem

Most people never change privacy settings. In most cases, the default is what happens.

This is well-documented in behavioral research, and it’s the reason AI companies that want your data design the default to include it. An opt-in default limits the training dataset to engaged, privacy-aware users. An opt-out default captures everyone who never changes settings — which is most people.

When a privacy policy changes to move training from opt-in to opt-out, the effective impact on training data volumes can be enormous. The change in wording is small. The change in practice is not.

The regulatory community has noticed. The FTC has said it may be unfair or deceptive for a company to adopt more permissive data practices through a retroactive amendment to existing users without meaningful notice. Under GDPR, a company that wants to use personal data for a new purpose — like training a model it didn’t train before — must notify users and, in many cases, get fresh consent.

Whether existing enforcement actions will catch up to the pace of these changes is an open question.


How to Check Your Own Apps

You don’t need to re-read every privacy policy you’ve ever agreed to, but a targeted audit is worthwhile for the apps that handle the most sensitive data.

Start with the apps that know the most about you:

  • AI assistants and chatbots (especially ones with memory features)
  • Voice input and dictation apps
  • Note-taking and journaling apps
  • Health tracking apps
  • File storage and document tools

For each one, look for:

  1. When was the privacy policy last updated? A recent update date warrants a read of the changes section.
  2. Is there a specific AI training section? Look for words like “improve our models,” “train our AI,” or “your content may be used.”
  3. What is the default? Does it say “unless you opt out” (opt-out default) or “only if you opt in”? These are very different things.
  4. Can you actually turn it off? Some policies describe training controls that require an Enterprise account, or that exist in the settings but are labeled confusingly.

If you use an AI dictation or keyboard app specifically, check the in-app settings for a “data and privacy” or “model training” section. Don’t rely solely on the public privacy policy — as the Wispr Flow situation shows, the policy and the help center can contradict each other.


When the Policy Changed, Not Just What It Says Now

One under-appreciated dimension of this problem is that historical training matters.

If an app trained on your data during a period when its policy said it wasn’t, or before you would have noticed the change, that data is already in the model. Deleting your account stops future exposure. It doesn’t remove what the model already learned.

This is why data minimization — not uploading personal content to services that may train on it — is a more robust strategy than managing settings after the fact. If your words, photos, voice notes, and documents aren’t on their servers, they can’t be used as training data regardless of what the policy says this month.


The Audit Pace Is the Signal

The fact that 25 of 58 AI privacy policies changed in a single month is itself the important finding.

It signals that this is an active, ongoing process — not a one-time decision that companies made when they launched. The policies will keep changing. The defaults will keep shifting. The only reliable protection is understanding which apps have access to your most personal data and deciding whether you want them to hold it at all.

For personal files, photos, and memories — the kind of content that feels most private — storage apps that don’t build AI products on user content exist. They have no competitive reason to add that capability, which means their policies don’t need to change.

Daftei stores personal files, photos, and memories privately. It doesn’t train AI models on user content, never sells data, and doesn’t run ads. The policy doesn’t change because the business model doesn’t depend on data. That’s the simplest answer to a problem that will otherwise require a monthly audit.

Your memories deserve better than an ad platform.

Try daftei free →
← All posts