What happens to your data if an AI company goes bankrupt?

What happened with Character.AI's bankruptcy and Disney?

Character.AI filed for Chapter 11 bankruptcy in Delaware in October 2026 and agreed to sell some of its assets to Disney for $20 million. Disney is the "stalking horse" bidder, which means it made the first offer and rivals can still beat it. That offer buys a non-exclusive, perpetual license to some of Character.AI's intellectual property, plus certain company claims. It doesn't buy the whole platform. A court still has to approve the sale, and the company expects the process to finish in late November or early December 2026. Until then the service keeps running. Users can open their chats and paid subscriptions stay active. If no buyer wants to keep the platform alive, the company says it will be wound down.

The collapse didn't come out of nowhere. Character.AI was one of the first breakout startups of the generative AI boom. In 2024 its founders and many staff left for Google in a deal reported at $2.7 billion, and the company stopped building its own large language models. After that came lawsuits over chatbot safety, rising costs and leadership departures. In September 2025, Disney itself sent a cease-and-desist letter over bots based on its characters. On user data, the company says customer information is still protected under its privacy policy. That same policy also says chat transcripts, voice recordings, images and videos may be disclosed as part of a sale or sale talks. Nobody has said publicly where those chats will end up.

Why should you care if you never role-played with a chatbot? Because the question is the same for any assistant that remembers your details, reads your contracts or checks your invoices. Your trust doesn't stay with the company you signed up with. It goes wherever the data goes. Here's my less popular view: most buyers spend ages comparing features and almost no time asking what happens if the company runs out of money. Yet that's the scenario that decides who reads your client notes in three years. A startup can be brilliant, well funded and still gone by next spring. The privacy policy you skimmed is the document that decides the outcome.

What happens to your chat logs and memories when an AI company fails?

When an AI company fails, your chat logs and memories are usually treated as assets of the business, not as your property. They can go to a buyer, stay with a restructured company, or be deleted when the service shuts down. Bankruptcy doesn't wipe a database. The servers, backups and files still exist, and a court decides what happens to everything the company owns. User data is often among the most valuable things left, especially for a company with millions of conversations. That's why the privacy policy matters so much. It's the closest thing you have to a contract about your data, and courts look at it when deciding whether the data can change hands.

US bankruptcy law has a specific safeguard here. If a company wants to sell personal information in a way its privacy policy doesn't clearly allow, the court can appoint a consumer privacy ombudsman to look at the deal. The 2025 bankruptcy of 23andMe, the genetic testing company, showed how this plays out. More than two dozen states objected to the company's first proposal. The ombudsman's report ran past 200 pages and recommended getting separate, affirmative consent from customers before their data was sold. One key finding: the privacy policy had been updated 22 times since 2007 and only mentioned bankruptcy from June 2022 onward. Millions of customers hadn't logged in since that change.

It helps to be specific about what "your data" covers, because it's more than the chats you remember. It includes the conversation history, any stored memories or profile facts, uploaded files like PDFs and spreadsheets, voice recordings, images, account details, payment records and usage logs. Then there's the part most people miss. If a company used your conversations to train its models, deleting the chats later doesn't necessarily remove what the model already learned. A wind-down isn't instant either. Backups can stay around for weeks or months after the app goes dark. So if a shutdown is announced and you think "good, it'll all disappear," don't count on it. Delete what you can while the delete button still works.

Can a new owner change the privacy policy after buying an AI company?

Yes, a new owner can usually change the privacy policy, but it's limited in what it can do with data that was already collected. The buyer generally takes on the promises the old company made. In the US, regulators have long held that making a material change to how existing data is used, without people's consent, can be a deceptive practice. Future changes usually just need notice. In the EU and UK, the buyer becomes the new data controller under GDPR. It needs its own lawful basis for using your data, and your rights to access and erase it come along with the data. That's real protection, but you have to use it yourself.

In practice, that protection often arrives as an email you never open. The new owner sends an "updates to our terms" message, your inbox files it with the promotions, and the change takes effect thirty days later. Before you trust any assistant, find the paragraph usually titled "business transfers" or "mergers and acquisitions." A typical line reads something like "we may share your information in connection with a merger, sale of assets, financing or bankruptcy." That sentence is the company telling you, in advance, that your data can be sold along with the furniture. Look closely at the wording. Some policies promise the buyer will honor the same terms. Others promise nothing at all.

Not every deal moves user data, and the Character.AI case shows why the details matter. Disney is licensing intellectual property, and the license is non-exclusive. As reported so far, it isn't a takeover of the user base. The harder question is what happens to everything else: the chats, the custom characters and the stored personas. They could go to another bidder or be wiped in a wind-down. Here's my take: for your privacy, an orderly shutdown with a clear deletion deadline is often better than a rushed sale to an unknown buyer. A shutdown ends the story. A sale to an unknown buyer starts a new one, with a privacy policy you've never read.

How do you check if an AI assistant is safe for client documents and invoices?

You check by asking four questions before you upload anything sensitive. Who owns the data? Can you export it? Can you delete it? And what does the policy say happens in a sale or bankruptcy? Start with ownership. The terms should say you keep ownership of what you upload and that the company only gets a license to process it so it can serve you. Then find the business-transfer clause and read it slowly. If it says data "may be transferred" with no promise that the buyer will honor existing terms, treat every client contract you upload as something a stranger might one day own. That doesn't mean you should never use the tool. It means you should be choosy about what you feed it.

Next, test export and deletion today, not when the headlines hit. Find the export option and actually run it, so you know what you get back and in what format. Then look for three controls: deleting a single chat, clearing stored memory, and deleting the whole account. Check how long the company says it keeps data after you delete it, because "we delete within 30 days" and "we may keep backups indefinitely" are very different promises. For an assistant with long-term memory, ask it straight out what it knows about you. If the answer surprises you, the memory feature is collecting more than you realized, and it's time to prune before you hand it a folder of invoices.

Finally, check two quieter risks: training and actions taken without asking. Does the company use your conversations to improve its models, and can you opt out? Can its staff read your chats for safety reviews? If the assistant can send emails or messages on your behalf, does it show you what it will send first and wait for your yes? That last question matters more every year as assistants move from answering to acting. For a fuller list of approval rules, see this checklist for trusting an AI assistant to act for you. If you're wondering whether to keep everything on your own machine instead, here's a plain comparison of local AI versus a cloud assistant for admin work.

How does this checklist apply to a WhatsApp assistant like Sonic?

You apply it the same way: read the policy, test what it remembers and decide what you're willing to send. Agent Sonic doesn't get a free pass because it feels familiar. The real difference with a WhatsApp assistant is visibility. Sonic keeps your memory and documents inside the chat you already use, so everything you've told it, and every answer it gave, sits in one thread you can scroll and search. There's no separate dashboard where your history piles up out of sight. You can even save a copy of the thread with WhatsApp's own export feature. That won't stop a company's servers from keeping data, but it does mean you're never guessing what you shared.

Start with memory, since that's where personal detail builds up fastest. Sonic remembers your preferences and context so you don't have to repeat yourself, and that's exactly why you should check what it has stored. Send "What do you remember about me?" and read the answer like an auditor would. Then think about documents. Forwarding a contract with "Summarize the risks and deadlines in this PDF" or an invoice with "Check this against last month's" is useful. But you can still black out bank details or client ID numbers if the summary doesn't need them. Before you upload anything regulated, read Sonic's privacy policy for retention, deletion and business-transfer wording. If something's unclear, ask the team directly.

The "does it act without asking" question matters most for outreach tasks, a Pro plan feature that's currently in Beta and may be paused from time to time. You say "Ask Dana when the invoice will be ready and let me know." Sonic shows you the exact first message, and nothing goes out until you reply "yes." It writes from its own WhatsApp number, introduces itself as your assistant and never pretends to be you. That's the kind of approval step the checklist is looking for. If you want an assistant that remembers what matters without you repeating it, and you can see that memory in a chat you already trust, leave your details to get Sonic's number and run the checklist yourself.

Frequently asked questions