As of late October 2026, we could source four incidents in which an AI companion or character chat service exposed user data: the Muah.ai breach (September 2024), a ten-minute Character.AI exposure (December 2024), an open server behind Chattee Chat and GiMe Chat (reported October 2025) and open cloud storage at Secret Desires (reported November 2025). A fifth incident, at the general-purpose app Chat & Ask AI in January 2026, is listed as related because it failed the same way.
Each entry below gives the date, what was exposed, and who reported it, with a link to the primary report. We add an incident only when the company itself, Have I Been Pwned, a named security research team or an established news outlet has reported it. This is not a ranking of which apps are "safe": a service missing from this list may simply never have been caught.
The list at a glance
| Reported | Service | What happened | What was exposed | Primary source |
|---|---|---|---|---|
| Oct 2024 (breach in Sept 2024) | Muah.ai | Breach of the site's data | 1.9 million email addresses, with AI prompts | Have I Been Pwned, 404 Media |
| Dec 2024 | Character.AI | About ten minutes during which parts of some accounts were visible to other users | Usernames, names, bios, characters, personas, voices or chats, for under 0.01% of users | Character.AI |
| Oct 2025 | Chattee Chat and GiMe Chat | An unprotected server left open on the internet | 43 million messages, 600,000+ images and videos, 400,000+ users, IP addresses and device IDs | Cybernews, via Malwarebytes |
| Nov 2025 | Secret Desires | Cloud storage containers left open | Nearly two million photos and videos, including photos of real women uploaded by users | 404 Media |
| Jan 2026 (related) | Chat & Ask AI | A misconfigured database | About 300 million messages from more than 25 million users, per the researcher | 404 Media |
Muah.ai, September 2024
The largest companion-app breach on record, and the one that showed what is at stake: prompts tied to real email addresses.
- When: Have I Been Pwned says that in September 2024 "the 'AI girlfriend' website Muah.AI suffered a data breach", and added it to its database on October 8, 2024. 404 Media first reported it the same day.
- Scale: 1.9 million email addresses. Accounts are not the same as people, so "1.9 million users" overstates it.
- What leaked: the exposed data includes AI prompts and email addresses. According to 404 Media, the prompts are linked to those addresses, many of which appear to be personal accounts with users' real names. 404 Media also reported that some of the prompts sought illegal content.
- Who can check: Have I Been Pwned flags it as a sensitive breach, so it isn't publicly searchable. Only the verified owner of an address can see whether it's in the data.
- What followed: on October 11, 2024, Malwarebytes reported that the data was "in use for active extortion attempts", adding that it was unknown whether these used platform activity or only the leaked email addresses. The Australian eSafety Commissioner's guide entry on the site describes Muah.AI as having "been the subject of significant data hacks".
We found no lawsuit or regulator action against Muah.ai in our searches, which doesn't mean none exists.
Character.AI, December 2024
A short exposure caused by the service itself, not an attacker, and disclosed by the company. In a blog post dated December 17, 2024, Character.AI said that for "approximately ten minutes", "less than 0.01% of our users had any information visible". What could be seen were parts of account pages: a username, name, bio, characters, homepage, personas, voices or chats. Most affected users had one section visible, and for chats, usually only the first page of one conversation with a single character. The company said it was revising how it changes critical infrastructure and adding technical controls.
Character.AI is back in the privacy conversation for a different reason in 2026. On October 12, 2026, it started a voluntary Chapter 11 process to sell the company under court supervision, and it says the platform could be wound down if no buyer keeps it running. Service and billing continue for now, and its FAQ says users can export their data or delete their account through the app or website. News reports say a group of state attorneys general has intervened over how user data is protected in any sale. The situation may change; our guide to what the bankruptcy means for your chats follows it, and saving your Character.AI characters and chats covers the export itself.
Chattee Chat and GiMe Chat, 2025
An open server, no hacking required. Researchers at Cybernews found that Imagime Interactive Limited, a Hong Kong company behind the two companion apps Chattee Chat and GiMe Chat, had left a streaming server (a Kafka broker) on the internet with no authentication. As Malwarebytes summarized on October 10, 2025, anyone who found it could see:
- 43 million messages between users and their AI companions
- more than 600,000 images and videos
- data on more than 400,000 users, along with their purchases
- IP addresses and unique device identifiers
Names and email addresses were not in the exposed data, but Malwarebytes noted that IP addresses and device IDs can be combined with other breaches to identify people. Chattee was the popular one of the two, with more than 300,000 downloads, mostly in the US. The leak was closed after the researchers disclosed it responsibly.
Secret Desires, November 2025
The incident that hurt people who never used the app. On November 19, 2025, 404 Media reported that Secret Desires, an adult chatbot and image platform, had left cloud storage containers open holding nearly two million photos and videos: both what users uploaded and what the platform generated. Many of the uploads were photos of real women taken from social media, some with their full names, which users had fed into the site's face-swap feature. According to the AI Incident Database's summary of the report, the files stopped being reachable shortly after 404 Media contacted the company.
The lesson for users is uncomfortable: on a platform that accepts photos of real people, a breach exposes people who never agreed to be there.
Related: Chat & Ask AI, January 2026
Not a companion app, but the same failure mode at a much bigger scale. On January 29, 2026, 404 Media reported that Chat & Ask AI, a popular AI chat app that claims more than 50 million users, had misconfigured its cloud database so that outsiders could reach user data. The independent researcher who found it said the database held about 300 million messages from more than 25 million users, including complete chat histories, timestamps, the names users gave the chatbot and how they had configured it.
What watchdogs and regulators have found
These aren't breaches, but they are the most cited findings about how companion apps handle data:
- Mozilla, February 14, 2024. Mozilla's Privacy Not Included team reviewed 11 romantic AI chatbots and gave every one its warning label; 10 of the 11 failed its minimum security standards. On Replika, it reported that the app records all the text, photos and videos users post, and that it accepted "11111111" as a password.
- Italy, 2025. Italy's data protection authority, the Garante, fined Luka Inc., the maker of Replika, €5 million in a decision dated April 10, 2025 and announced May 19, 2025, finding no legal basis for its processing as of February 2023, an inadequate privacy policy and no age verification at the time. It opened a separate investigation into how the model was trained.
- United States, September 11, 2025. The FTC sent orders to seven companies, Character.AI's maker among them, to study AI chatbots that act as companions.
What the incidents have in common
None of them needed a sophisticated attack. Three were infrastructure left open to the internet (a streaming server, storage containers, a database with permissive rules), one was a deployment mistake by the company itself, and the one with an attacker involved a site its own hacker described to 404 Media as "basically a handful of open-source projects duct-taped together".
And the damage comes from the content. A leaked password can be changed; a leaked roleplay prompt tied to your email address can't be taken back. That's why what a service stores, and for how long, matters more than any promise about security.
What you can do
- Check your address on Have I Been Pwned. For sensitive breaches like Muah.ai's, you have to subscribe and verify your address to see the result.
- Use a separate email for companion and roleplay apps, without your real name in it. The Muah.ai data was damaging because prompts were tied to personal addresses.
- Keep identifying details out of chats and personas: full name, employer, city, photos of yourself.
- Never upload photos of real people. It is banned on most platforms, and Secret Desires showed who gets hurt when that data leaks.
- Delete chats and accounts you no longer use, and read the retention clause: some sites keep data for years after you leave. Our guide to what AI roleplay sites store, and for how long compares them.
- Use a unique password and two-factor sign-in where offered, as Malwarebytes advises after every breach.
- If someone threatens to expose your chats, don't pay and don't reply. Keep the message, verify any "official" contact through a separate channel, and report it to the police (in the US, to the FBI's Internet Crime Complaint Center).
- Prefer setups that keep your characters on your own device. Less data on a server means less data in a breach.
How Goony stores your data
Goony is our product, so here is how it works rather than a promise. Characters you create or import locally, and your Offline chats, live in your browser, not in Goony's database. Each reply is still generated on Goony's servers. Cloud chats, the default when you're signed in, are stored on Goony's servers, and Goony's Privacy Policy says requests to the AI models are logged for as long as the account exists. If you'd rather a character not be stored in Goony's database, import it as Local and chat Offline; each reply still passes through Goony's servers, and model requests are logged as the policy describes. The rest of the product is explained in what Goony is and how it works.
FAQ
Has Character.AI had a data breach?
Character.AI disclosed one incident: in December 2024, for about ten minutes, parts of some accounts (including chats in some cases) were visible to other users, affecting under 0.01% of users. We found no report of an outside attack on it.
Was Muah.ai hacked?
Yes. Have I Been Pwned says Muah.ai suffered a breach in September 2024 that exposed 1.9 million email addresses along with AI prompts, and Malwarebytes later reported extortion attempts using the data.
How do I know if my AI companion data leaked?
Check your email address on Have I Been Pwned; for sensitive breaches you need to verify the address first. Also watch for notices from the app, and treat unexpected emails that mention your chats as possible extortion or phishing.
Which AI companion apps have leaked chats?
Among the incidents we could source: Muah.ai (2024), Chattee Chat and GiMe Chat (2025), and briefly Character.AI (2024). Secret Desires exposed photos and videos in 2025, and the general AI chat app Chat & Ask AI exposed messages in 2026.
Can someone blackmail me with leaked AI chats?
It has been attempted: after the Muah.ai breach, Malwarebytes reported active extortion attempts. Don't pay or reply, keep the evidence, and report it to the police.
Is a chat app that stores data locally safer?
Data that isn't on a server can't leak from that server, so local storage shrinks the risk. Most hosted apps still process each message on their servers to write a reply, so check what they log.




