100,000+ ChatGPT logins hit the dark web, OpenAI says it wasn't hacked

People often paste private thoughts, internal roadmaps, and proprietary code into ChatGPT, assuming it sits safely behind OpenAI's enterprise-grade defenses. But bad actors didn't need to breach OpenAI to see what you've been working on. They just harvested it directly from everyday laptops.

A threat intelligence report from Singapore-based cybersecurity firm Group-IB revealed that over 100,000 compromised credentials containing ChatGPT accounts were listed on dark web marketplaces between June 2022 and May 2023. The surge tracked the chatbot's parabolic growth, hitting an all-time peak of 26,802 compromised logs in May 2023 alone, with the Asia-Pacific region seeing the highest concentration of stolen data.

Stealers over zero-days

OpenAI quickly moved to clarify that its own servers were never compromised. A spokesperson confirmed the data dump was the result of commodity malware on users' devices rather than an OpenAI breach, adding that the company maintains industry best practices for authentication and authorization.

In plain English, hackers didn't break into the mothership. They used off-the-shelf information-stealing malware to scrape saved browser sessions, cookies, and auto-filled passwords straight off compromised personal computers.

A single stealer family called Raccoon accounted for a massive 78,348 stolen ChatGPT accounts. Vidar trailed in second place with 12,984 accounts, followed by RedLine with 6,773. Raccoon operates on an accessible subscription-as-a-service model on the dark web, meaning virtually anyone with bad intentions and a small budget can launch attacks without writing custom exploits.

The goldmine in your history tab

Losing an account login is annoying, but losing access to ChatGPT is fundamentally different from a forum credential leak. Because ChatGPT retains conversation history by default, anyone buying these logs gets direct access to entire chat logs.

Those sessions contain an alarming amount of proprietary data. Employees regularly use ChatGPT to refactor proprietary code, summarize non-disclosure agreements, and draft quarterly roadmaps. When combined with personal diaries and credentials for other accounts captured in the same malware payload, it hands bad actors an unvarnished blueprint of internal business operations.

Why it matters

This dump is a stark reminder that generative AI tools have quietly become one of the biggest insider-threat vectors in tech. If you use ChatGPT for work, enabling two-factor authentication and auditing your saved chat history is no longer optional. It is worth doing right now.

For enterprise security leaders, this clarifies why companies like Samsung moved aggressively to restrict generative AI tools on company hardware. The threat model is no longer just whether OpenAI trains on corporate prompts; it is that end-user machines are constantly vulnerable to commodity info stealers.

We think OpenAI's technical response is accurate, but pointing fingers at endpoint hygiene only solves part of the equation. As chatbots handle increasingly sensitive tasks, AI platforms will need more aggressive default session expirations and sensitive-data redaction. Relying on users to pick strong passwords simply does not cut it when their chat logs read like corporate wiretaps.