OpenAI Forum Flaw Exposed Employee ChatGPT Accounts and Internal GitHub Code

Researchers exploited a configuration flaw in OpenAI’s community forum, which is hosted by the third-party platform Discourse. The vulnerability gave them access to internal sign-on systems and, ultimately, to OpenAI employees’ ChatGPT accounts.

One of those ChatGPT accounts had access to internal code through GitHub, raising concerns about the security risks of connected AI accounts and third-party platforms.

OpenAI says the forum security issue has been fixed

“We would like to thank the researchers who contacted us and shared their findings,” OpenAI said, adding that the issue has been fixed.

Anthropic declined to comment. Hactron did not immediately respond.

Anthropic reports rapid growth in AI-assisted research

The revelation Thursday, first reported by the Wall Street Journal, came as Anthropic published new data showing the company’s rapidly increasing use of AI to develop new models.

Anthropic said that 26% of its research and development work was “led” by Claude models, up from 1% in March. The company said this means AI completed most of the tasks under human direction and supervision.

As AI systems become more powerful, Anthropic said they are “increasingly being used to build the next version of AI systems.”

Anthropic examines the possibility of recursive self-improvement

Anthropic said it shared the data to help the public understand “how close the world is to recursive self-improvement”—the point at which AI can train and improve itself or create new models.

This threshold is central to concerns that AI systems could become more difficult to monitor, potentially increasing the risk of losing human control.

Anthropic added that none of the studies examined showed models operating fully autonomously. In 90% of tasks, AI “collaborates” with humans to perform most of the work.

© 2026 Financial Times Company. Unauthorized reproduction is prohibited. May not be redistributed, copied or modified in any way.