How to Turn Off the AI You Never Asked For

→ The Pattern — Instagram's May 8 deadline and the WhatsApp lawsuit
→ The 15-Minute Reset — five platforms, five switches, in order
→ The Full Reset Guide — everything that didn't fit in the email, free
→ Norton Neo — the sponsor, reviewed honestly
→ One Thing I Noticed — two reader quotes that tell the whole story

You opened an app this week and something was different. You didn't change anything — but something changed on you.

That's not paranoia. It happened.

444 readers voted in this week's poll. 79% said the same thing in two different ways: they didn't know AI features had been added to their apps, or they did know and found it unsettling. Dozens wrote in and asked one question directly: can I turn it off?

Give me 15 minutes. Five platforms. Every switch that matters.

Every major platform — Meta, Apple, Google, Chrome — added AI features to apps you already use, turned them on by default, and didn't ask.

Most urgent: Instagram ends encrypted DMs on May 8 — after that, Meta can read your messages. If you turned on encrypted chats, download your history before then.

A class-action lawsuit filed in January in federal court in California alleges WhatsApp's privacy claims don't cover how messages are stored. Nothing proven — but privacy groups are watching.

If you followed Tuesday's guide and turned this on: the full picture is that transcription runs entirely on your device. Your audio never leaves your phone. Tuesday's recommendation stands — this is about making sure you know where the control lives.

The switch, if you want it off:

iPhone users: Siri must be enabled for this feature to function. Turning off Siri also disables transcription.

Leaving it on is a legitimate choice. Turning it off is too.

WhatsApp voice transcription uses a speech recognition model running locally on the iPhone's dedicated AI chip. The audio file is never transmitted to any server. What is transmitted is the text transcript if you copy and paste it into a chat — at that point it becomes a message subject to standard WhatsApp server processing. On-device is not the same as fully private. It's meaningfully better than cloud-based transcription.

Meta AI appears in your search bar, summarizes comments on your posts, and uses your interactions to inform its models. There is no master off switch.

Mute on Instagram:

Mute on Facebook:

To remove the Meta AI ring from your search bar entirely:

On training objections: If you're in the EU or UK, Meta's Privacy Center has a "Right to object" form under AI at Meta. US readers: this option doesn't exist for you under current federal law. Muting plus not engaging with Meta AI is your only lever.

The EU/UK "Right to object" form invokes Article 21 of Europe's data protection law (GDPR) — the right to object to how a company processes your personal data. Meta is legally required to respond within 30 days and must demonstrate compelling legitimate grounds to override the objection. In practice, Meta's response typically offers limited processing adjustments rather than full cessation. The form is not an off switch. It's a legal record that your objection exists.

Additionally: Meta's training data policy distinguishes between public posts (used for training by default, globally) and private messages (not used for training, per Meta's stated policy). The class-action suit in The Pattern section challenges whether that distinction holds in practice.

Apple Intelligence is on by default on supported iPhones and learns from every app unless you limit it.

Per-app control:

To turn it off entirely:

Turning this off stops new data use. It doesn't address data Apple collected before you flipped the switch. The full guide below covers what can't be undone.

The settings menu is labeled "Apple Intelligence & Siri" on all iOS 18+ devices.

Apple's on-device AI processing handles most tasks locally without sending data to Apple's servers. The exception is Private Cloud Compute — a system Apple uses for more complex requests that can't be handled on-device. Apple claims PCC requests are processed without Apple being able to access the data, and they've published the architecture for external review. That framework has not been publicly contested by security researchers to date, though not every implementation detail has been independently verified. "On-device" and "Private Cloud Compute" are two different things. Most Apple Intelligence tasks use on-device. Some don't.

Turning off Apple Intelligence under Settings → Apple Intelligence & Siri stops both. It does not delete data already processed before the toggle.

Google is replacing Google Assistant with Gemini — its new AI assistant — as the default across Android. "Hey Google" may now trigger Gemini without you having switched.

To switch back:

Samsung-specific:

If "Digital assistants from Google" doesn't appear, search "Gemini" in system settings.

Google Assistant routes voice queries to Google's servers and returns results. Gemini does the same, with one addition: Gemini can access and act on content within apps — Gmail, Google Calendar, Google Docs — via cross-app context. This means Gemini can read the email you're currently drafting to help you finish it. That's functionality Google Assistant never had, and it's the capability gap that makes switching back consequential, not just preferential.

The switch-back path restores Google Assistant as the default invocation trigger. It does not disable Gemini from operating in individual Google apps where it's been separately enabled.

Chrome added AI-powered writing help, a Gemini sidebar, and browsing-behavior suggestions. Most are on by default — and they're not where you'd expect to find them.

Standard path:

Deeper control:

The flags path is for readers who want full granularity. The Experimental AI settings path works for most people.

The two relevant flags:

chrome://flags/#prompt-api-for-gemini-nano — controls whether websites can access Gemini Nano directly via the browser's Prompt API. This allows third-party web apps to run AI inference using Chrome's built-in model. Disabling this prevents that cross-site access.

chrome://flags/#optimization-guide-on-device-model — controls whether Chrome downloads the Gemini Nano model to your device at all. Set to Disabled and relaunch to prevent the model from being stored locally.

Both flags confirmed active on Chrome 146.0.7680.177+. Flag names occasionally change between major versions — if one doesn't appear, search "Gemini" in the flags search bar to find the current equivalent.

One reader — a Harvard researcher — wrote: "I shouldn't have to police my own devices." Another just asked: "I would like to know if I can turn it off anywhere." Those two messages together tell the whole story. This issue exists for both of them.

If you're in the EU or UK, you have a formal option US readers currently don't: the right to object to Meta using your data for AI training.

What this does: It signals your objection to Meta using your past and future interactions for AI model training. Meta is legally obligated under Europe's data protection law (GDPR) to respond within 30 days. It does not disable Meta AI in your apps — you still need the mute steps from the email.

The honest framing: Submitting the form starts a process. It is not the same as Meta stopping. Track whether you receive a confirmation within 30 days.

US readers: This option is not available under current federal law. That may change. It hasn't yet.

💬 Join the conversation →

What's changing: Meta has officially announced it will end support for end-to-end encrypted DMs on Instagram on May 8, 2026. End-to-end encryption meant only you and the person you were messaging could read the conversation. After May 8, DMs revert to standard encryption, which means Meta can access message content. Instagram has cited low adoption as the reason.

What to do before May 8:

Instagram is actively prompting users to download their histories now. Don't wait for the deadline.

Not all messages are encrypted — only conversations where both parties explicitly turned on encrypted chats. If you never turned this on, nothing changes for you.

💬 Join the conversation →

What the lawsuit claims: Filed in federal court in California, the suit alleges that WhatsApp's end-to-end encryption claim contains a structural gap the marketing doesn't disclose. The central allegation: messages backed up to Google Drive or iCloud are not end-to-end encrypted. Those backups are accessible to Meta and potentially third-party contractors. The suit argues this creates a consumer fraud — the headline privacy promise is technically true for messages in transit but misleading about how those messages are ultimately stored.

What's proven: Nothing. The case is in early stages.

Why it matters anyway: The allegations are specific enough to draw attention from privacy advocacy groups including EFF and Privacy International. Regardless of how the lawsuit resolves, if you back up WhatsApp to iCloud or Google Drive, your messages may be less protected than the marketing suggests.

You can close this gap:

Read the original coverage and form your own view.

💬 Join the conversation →

The structural gap the lawsuit identifies: WhatsApp messages are end-to-end encrypted in transit between devices. When you enable iCloud or Google Drive backup, those backed-up messages are stored in the cloud without end-to-end encryption. Apple and Google hold the decryption keys for their respective cloud services. This means a government subpoena to Apple or Google can produce your WhatsApp message history even though a subpoena to WhatsApp directly would yield nothing readable. This is a known, documented gap — not a new allegation. The lawsuit's claim is specifically that WhatsApp's marketing presents the encryption as more comprehensive than it is given this backup behavior.

This didn't fit the email — too technical for the format. But if you want full Chrome hygiene:

Chrome recently patched its fourth actively exploited security flaw of 2026 — tracked as CVE-2026-5281 (a standardized security vulnerability ID). It affects the part of Chrome that runs website code, and it was being used against real users before a fix existed. A flaw discovered and exploited before a patch is available is called a zero-day.

Check your Chrome version:

This is separate from the AI settings — it's a security patch. Both matter. Do both.

Version confirmed current as of April 17, 2026.

💬 Join the conversation →

Some things in this guide have no real opt-out. Naming them directly:

Meta AI in WhatsApp: You can mute it. You cannot turn it off. Meta AI is built into the WhatsApp platform. If you use WhatsApp, you share a platform with it. The mute step reduces your surface area — it doesn't eliminate it. If this concerns you, use a different app for sensitive conversations. Signal is the most widely cited alternative.

WhatsApp cloud backups: If you back up WhatsApp to iCloud or Google Drive, those backups are not end-to-end encrypted by default. You can close this gap under:

Meta training on your public posts: If you've ever posted publicly on Facebook or Instagram, that content has likely already been used in Meta's AI training. The objection form (EU/UK only) addresses future use — it doesn't retroactively remove what's already been used. This is true of most major platforms.

Apple Intelligence and prior Siri data: Turning off Apple Intelligence stops new collection. It doesn't erase data Apple gathered from prior Siri interactions and app usage before you flipped the switch.

The honest summary: You have more control than you had yesterday. You don't have full control. Anyone claiming you do is selling something.

💬 Join the conversation →