AI companion data breaches: known incidents and how to protect yourself

Companion chats contain some of the most private things people write: feelings, fantasies, health worries, details about partners. When a companion service is breached, that text can be tied back to a person and used for embarrassment or extortion.

This page lists publicly documented incidents with dates and sources: breach notification services, the reporting security researchers, and statements from the companies where they exist. It is not a complete record; many incidents are never disclosed. Checked October 23, 2026. For the wider privacy picture, read the AI companion privacy guide.

Date	Service	What was exposed	Source
September 2024	Muah.AI (companion site)	1.9 million email addresses with users' prompts	Have I Been Pwned
Reported October 2025	Chattee Chat and GiMe Chat (companion apps by Imagime Interactive, Hong Kong)	About 43 million messages and over 600,000 images and videos from over 400,000 users, with IP addresses and device IDs	Cybernews, Malwarebytes
Reported February 2026	Chat & Ask AI (general AI chat app by Codeway, not a companion app)	About 300 million messages from about 25 million users	Malwarebytes
March 2026	17 unnamed companion apps on Google Play (security audit, not a breach)	Flaws that could give attackers a route to stored chats in 10 of the 17 apps	Oversecured, SecurityBrief

Details of each follow.

Have I Been Pwned (HIBP) records the breach date as September 17, 2024 and added it on October 8, 2024: 1,910,261 email addresses, AI prompts and sexual preferences. HIBP classifies it as a sensitive breach, so it does not appear in public searches; only the verified owner of an address can see whether it was included. The tech news site 404 Media first reported it. According to Malwarebytes, the site's administrator claimed the attack was sponsored by competitors, and extortion attempts followed. The law firm Linklaters described reports of criminals asking IT employees for access to their employers' systems.

We do not recommend Muah.AI. Full details and steps for affected users are in Is Muah AI safe?

Cybernews researchers found that the developer, Imagime Interactive Limited, had left a Kafka Broker server (a system for streaming data between services) open to the internet with no authentication. Anyone who found it could watch chats between users and their AI companions in real time, along with links to photos, videos and generated images. Malwarebytes, summarizing the research in October 2025, reported 43 million messages, more than 600,000 images and videos, and more than 400,000 users across iOS and Android. No names or email addresses were found, but IP addresses and unique device identifiers were. The exposure was closed after responsible disclosure.

The lesson: a breach does not need a hacker. A setting left open is enough.

This is a general-purpose AI chat app, not a companion app, but it shows how large chat logs get exposed. According to Malwarebytes, an independent researcher found a database belonging to Chat & Ask AI, developed by Codeway, readable without authentication because of a Firebase misconfiguration. The researcher reported access to about 300 million messages from about 25 million users. Codeway fixed the issue within hours of being notified. The same researcher's scan found similar misconfigurations in 103 of 200 iOS apps checked.

In March 2026, the mobile security firm Oversecured published an audit of AI companion apps on Google Play. As reported by SecurityBrief, it found 14 critical and 311 high-severity issues across 17 apps, and in 10 of them a possible route for attackers to reach users' stored conversations. Problems included developer keys shipped inside the app and a flaw that let other apps on the phone inject scripts into a chat. The researchers did not name the apps because the flaws were unpatched at publication.

This was an audit, not a confirmed breach, but it fits an older finding: Mozilla's 2024 review said 10 of 11 romantic chatbots failed its minimum security standards (Mozilla Foundation).

• Misconfiguration, not genius hacking. Open servers and public databases caused the largest exposures.
• Chats linked to identity. An email address, IP address or device ID is enough to tie intimate text to a person.
• Media makes it worse. Uploaded photos and generated images were exposed along with text.
• Small developers, big audiences. Several apps with hundreds of thousands of users came from small teams without visible security practices.
• Silence from operators. In most cases the public learned about the problem from researchers, not from the company.

1. Use a separate email address with no real name, used only for companion services.
2. Keep identifying details out of chats: your name, address, workplace, and anything about real people.
3. Do not upload photos of yourself or anyone else. Several services ban uploads of real people's photos for this and other reasons.
4. Read the privacy policy for retention periods, training use and a named operator. The privacy guide shows what to look for.
5. Search the service name with "breach" before you pay.
6. Sign up for breach alerts with Have I Been Pwned's Notify Me service, which also covers sensitive breaches.
7. Use unique passwords and two-factor authentication on your email.
8. Delete accounts you no longer use, after cancelling payments. The billing guide explains cancellation.

If someone contacts you with leaked data: do not pay, keep the messages, and report it. In the US that means the FBI's Internet Crime Complaint Center.

To compare services on their published privacy and safety rules, see the companions category.

• Have I Been Pwned: Muah.AI
• Malwarebytes: Muah.AI breach (October 2024)
• Linklaters (October 14, 2024)
• Cybernews: Chattee Chat and GiMe Chat
• Malwarebytes: Chattee Chat and GiMe Chat (October 2025)
• Malwarebytes: Chat & Ask AI (February 2026)
• Oversecured audit (March 23, 2026)
• SecurityBrief on the Oversecured audit
• Mozilla Foundation (February 2024)

Checked October 23, 2026.