A misconfigured storage bucket at 302.ai left 17.6 million files open to anyone who found it, according to research from Cybernews. Among them were more than 144,000 AI chat logs containing personal information. The 302.ai data leak of AI chat logs is a pointed reminder that what you type into an AI tool is only as private as the storage behind it.
Reporting on the incident says no malicious access was confirmed. That is reassuring, but it does not mean the files were safe in any meaningful sense, as we explain below.
What Cybernews found in the 302.ai bucket
Cybernews researchers discovered that a storage bucket tied to 302.ai was publicly reachable. The exposure covered 17.6 million files, and the headline coverage describes the total as 22.7TB of data. Inside were more than 144,000 AI chat logs that included personal information. Coverage of the story also points to support logs related to an adult product, which illustrates how intimate some of the material can be.
The key detail is how the exposure happened. This was not described as a sophisticated intrusion. A storage bucket that was open to the internet meant that anyone who located it could view the contents, with no hacking required. According to the reporting, there is no confirmation that a malicious party accessed the data.
The source material does not give further technical detail, so it is not clear how long the bucket was open or whether every file contained sensitive content. What is reported is the scale and the presence of personal information in chat logs.
Why exposed AI chat logs are a privacy risk
People tend to talk to AI tools differently than they talk to search engines. They paste in work documents, describe health worries, ask for relationship advice, and share details they would hesitate to put in an email. When those conversations are stored in plain view, the content can reveal a lot about a person.
A few reasons chat logs deserve extra care:
- They are written in natural language. Names, addresses, and personal circumstances often appear in context, which makes them easy to read and hard to redact.
- They can be linked to a person. Even without a name, details inside a conversation can point to who wrote it.
- They are rarely deleted. Many platforms keep conversation history for long periods, so an exposure can reach back well beyond recent activity.
- Support logs add a second layer. Customer service records can tie a chat to an order, an account, or a product, which may be embarrassing or sensitive.
"No confirmed malicious access" is also a limited assurance. Unless a provider has complete access logs for the period the bucket was open, it can be difficult to prove that nobody else looked. Automated scanners constantly search for open buckets, so exposed data is often discovered quickly.
How often AI platforms leak data through misconfiguration
The 302.ai case fits a pattern of AI-related services exposing data through setup mistakes rather than clever attacks. The Vibetech case is a recent example: an unsecured repository left millions of records open to anyone with an internet connection, as covered in our report on the Vibetech database exposing 13.7M records.
AI companies can also be hit by conventional breaches. Our coverage of the Suno breach affecting 55M users shows how quickly a fast-growing AI platform can accumulate large amounts of user data, and how much of it can be exposed at once.
The common thread is speed. AI products are often built and scaled quickly, and storage that is meant to be private can end up public through a single wrong setting. Researchers continue to document how flawed AI tooling can leak data, as discussed in our piece on AI ransomware and flawed test models leaking data.
What This Means For You
You cannot inspect how an AI provider configures its storage, so you have to assume that anything you submit could one day be exposed. That does not mean avoiding AI tools. It means being deliberate about what you share with them.
If you have used a lesser-known AI platform, or one that resells access to other models, you may have less visibility into who stores your data and for how long. A VPN does not help here: it protects your connection in transit, but it cannot protect data that a provider stores on its own servers.
How to reduce what AI tools know about you
You can lower your exposure with a few habits:
- Keep sensitive details out of prompts. Avoid pasting passwords, financial data, medical information, ID numbers, or confidential work documents.
- Use placeholders. Replace real names and addresses with generic stand-ins when asking for help drafting or editing.
- Review retention settings. Check whether the service saves history, and turn it off or delete old conversations where the option exists.
- Delete what you no longer need. Remove old chats and uploaded files, and close accounts you no longer use.
- Use unique passwords and two-factor authentication. If an account is exposed, this limits the damage to other services.
- Watch for follow-up scams. Leaked personal details can fuel convincing phishing messages, so treat unexpected requests with suspicion.
Key takeaways
The 302.ai data leak of AI chat logs shows that a single storage setting can expose millions of files, even when no attacker is confirmed. Treat AI chats as potentially permanent records, share less, and clear out history you do not need. For more on how AI services handle and mishandle user data, read our coverage of the Vibetech database exposure and the Suno breach.
