Security researchers say they used Anthropic's Claude chatbot while probing OpenAI's systems, in an operation the company later acknowledged as a sanctioned test.

The team, from startup Hacktron AI, accessed ChatGPT accounts belonging to OpenAI staff by way of an internal discussion forum. They then submitted a harmless code-change request to OpenAI's GitHub repository.

Hacktron reported the findings through OpenAI's bug bounty programme and received a $6,500 payment. The researchers said they had access to repository code but did not download it.

An OpenAI spokesperson thanked the researchers and said the exploited vulnerabilities had been addressed. Hacktron added that AI tools have sharply cut the time needed to plan and carry out such attacks.

Source: The Guardian