Is Meta AI safe with your data after the $25B ruling?

What did the $25 billion Meta ruling actually decide?

On October 21, 2026, a New Mexico judge ordered Meta to pay $25 billion after a jury found that the company misled Facebook users about its privacy practices. The case grew out of the Cambridge Analytica scandal, and the award is one of the largest penalties ever imposed on a tech company. Judge Francis Mathew of the First Judicial District in Santa Fe set the amount about a month after the September jury verdict. The jury had found that Meta violated the state's Unfair Practices Act more than 43 million times. New Mexico filed the suit back in 2021, and it was the only state to take its claims all the way to trial. Forty-eight others settled. Meta has said it will appeal, so no money will change hands for a while.

The penalty could have been much bigger. Each violation carries a maximum penalty of $5,000. Across 43,899,720 violations, that adds up to about $219.5 billion. The judge cut it to $25 billion. The state had asked for between $35 billion and $45 billion. For comparison, the Federal Trade Commission's 2019 penalty over the same scandal was $5 billion, and in 2022 Meta paid $725 million to settle a class action that grew out of it. Attorney General Raúl Torrez has said the money will go into a fund for the state's education system. One nuance matters here. The case wasn't really about the data Cambridge Analytica harvested. It was about what Facebook told New Mexico residents about how their information was handled.

So what does a courtroom in Santa Fe have to do with the assistant in your WhatsApp? Quite a lot, if you run your business from your phone. At its core, the jury's finding was about trust: the gap between what a company says it does with your data and what it actually does. You face that same question every time an AI assistant asks to read your chats, your PDFs or your calendar. Here's our take. The fine changes nothing on your phone tonight. Meta's settings, defaults and policies are the same today as they were last week. What the ruling gives you is a good reason to stop tapping 'Allow' on autopilot. That goes for Meta's assistants and for everyone else's, ours included.

Does the ruling change how Meta AI uses your WhatsApp data?

No. The ruling punishes Facebook's past conduct. It doesn't change how Meta AI handles what you type into WhatsApp today, and today's rules deserve a careful read. Since December 16, 2025, Meta has used people's conversations with Meta AI, typed or spoken, to personalize the content and ads they see across its apps. If you use Meta AI, there's no switch to opt out of that. Meta says it won't use sensitive topics such as religion, health, political views, sexual orientation or ethnic origin to target ads. The rollout also initially left out the European Union, the UK and South Korea. If you live anywhere else, assume that what you tell Meta AI helps shape the ads you see on Facebook and Instagram.

There's an important line here, and it's easy to blur. Your ordinary WhatsApp chats with your accountant, your clients or your mum are end-to-end encrypted, and the ad policy doesn't mean Meta is reading them. The change covers your conversations with Meta AI itself. When you ask it to draft a reply, sum up a long message or plan your week, that exchange goes to Meta, and that's the part that can feed recommendations. Meta also says your WhatsApp AI chats won't personalize your experience on its other apps unless WhatsApp is linked to the same Accounts Center. The rule of thumb is simple: anything you paste into Meta AI is something Meta now knows about you.

Can you stop Meta AI from training on your data? Only partly, and it depends on where you live. In the EU, Meta has offered an objection process, though one privacy watchdog complained in 2025 that people who had already objected were asked to do it again. Elsewhere you have fewer options. Keep client details and contracts out of Meta AI, check what your Accounts Center links together, and review your ad preferences. Those change what you see, not whether your data is processed. Business owners have a second front: Meta's commercial terms cover crawling websites linked to business pages, so here's how to block Meta's AI crawler from your site. The wording has shifted since then, so check what Meta's revised terms still allow before you assume you're covered.

Is Meta Muse safe to use for your work?

Muse is the most privacy-conscious assistant Meta has built so far, but safer than before isn't the same as private. Muse launched in late September 2026 as a personal AI agent that can act across the apps you connect to it. Meta gives each user their own isolated cloud virtual machine and runs a supervisory layer on top, which it calls the Sentinel agent. Sentinel manages permissions and flags sensitive tasks before they run. Meta has also said Muse data won't feed its advertising systems. Later in 2026 it plans to test a stricter Confidential VM, where your data would be encrypted with keys only you hold. Those are real design choices, and they deserve credit. They're also not the whole story.

The catch is in the fine print. In the current setup, Meta staff can still reach your data to support, secure or run the service. Meta's own wording is that its design restricts staff access through operational policies. That's a promise, not a lock. Reports also disagree about whether the setting that lets Meta use records of what Muse does for you to improve its models starts on or off. Meta calls those records sanitized, but you still have to decide whether you're comfortable with them leaving your account. The disagreement itself is the lesson. Don't trust a headline about defaults, including ours. Open the settings and check them yourself, and check again after every major app update.

If you use Muse, spend five minutes on four things. In the data controls, turn off the option that helps improve Meta's AI models. Meta says that change also covers your earlier interactions. In permissions, switch to having Muse ask before every action, so nothing gets sent or booked without a tap from you. In connectors, disconnect every app Muse doesn't need this week. You can always reconnect it later. Then ask Muse to forget any personal details you didn't mean to share. If Muse isn't available in your country yet, or you'd rather not add another Meta account to your life, there are WhatsApp assistants you can use instead of waiting for Muse.

What should be on your AI assistant privacy checklist?

A good AI assistant privacy checklist asks five questions. What can it read? What can it do without asking? Does it train on your data? Who else can see it? What can you switch off or delete? Ask them of every assistant, Meta's or anyone else's, before you connect your chats, documents or calendar. If you run a business, write the answers down. 'I think it's fine' won't help much when a client asks where their contract went. Start with the vendor's own help pages, then test the claims: look in the settings, try deleting something and see what's left. Five minutes now beats an awkward email to a client later, and the checklist works just as well on a tool you already use as on one you're still considering.

Reading and acting are the two questions that cause the most trouble. An assistant that only sees what you send it is very different from one that can open your whole inbox. An assistant that drafts a message for your approval is safer than one that sends replies on its own. Ask directly: if it messages someone for me, does it say it's an assistant, and do I see the text first? Satya Nadella has said to assume any AI model could be compromised. That sounds dramatic, but it's sensible planning, and our plain-English checklist for handing tasks to an AI covers which jobs should need your yes every time.

Training and off switches are where vague language hides. 'We use your data to improve our services' can mean almost anything. You want plain answers: are my conversations used to train models, and can I turn that off? Then think about what happens when you leave. Can you delete your history, wipe what the assistant remembers about you and disconnect your calendar without emailing support? One contrarian point: don't choose an assistant based on who got fined this week. A big fine tells you what a company did years ago. Its current settings and defaults, and how it answers a direct question, tell you what it does now. A small company with vague answers is no safer than a giant with a court record.

How does a WhatsApp assistant like Sonic stack up on the same questions?

Judge Agent Sonic the same way you'd judge Meta: by what it can read, what it can do and what you control. Sonic is an AI assistant you message in WhatsApp like any other contact, with nothing to install. It works with what you send it: a voice note, a forwarded PDF, a contract, an invoice, or a request like 'Remind me to call the accountant every Monday at 9.' It only uses your Google Calendar, Sheets or Docs if you connect them, and its Notion connection is read-only. In a group, Sonic is there because someone added it, and that raises consent questions for everyone else in the chat. We've written up four simple consent rules for AI in group chats.

The 'can it act without asking' question matters most for outreach tasks. That's Sonic's Pro feature for messaging people on your behalf. It's in Beta, so it may be paused from time to time. You say, 'Ask Dana when the invoice will be ready and let me know.' Before anything goes out, Sonic shows you the exact first message, and nothing is sent until you reply 'yes.' It writes from its own number, introduces itself as your assistant and never pretends to be you. It runs one task at a time for up to three people a day, and it only sends first messages between 08:00 and 21:00 in each person's time zone. If someone replies 'stop,' Sonic never messages them again. The step-by-step outreach tutorial shows the whole flow.

Don't take our word for the rest. Ask us what you'd ask Meta: do you train on my conversations, who on your team can see them, and how do I delete what you remember about me? You should get a straight answer in writing before you forward a single contract. That's the standard the New Mexico case should set for every assistant, and we'd rather be held to it than let off the hook. If the answers work for you, get access to Sonic. Leave your details and you'll get Sonic's number and onboarding. Then start small, with a reminder or a to-do list. Hand over more only as it earns your trust, because that's how delegation should work with any assistant.

Frequently asked questions