**OpenAI** fired three researchers, saying they mishandled sensitive company information, including work with external organizations related to evaluating AI models.
Key points
OpenAI fired three researchers over allegations that it violated sensitive information management regulations.
At least two were reportedly involved in research related to AI safety.
The measures were taken at a time when security incidents involving recent agents are intensifying the safety debate.
Background behind the firing of OpenAI researchers
According to a BBC report, OpenAI, the company that develops ChatGPT, did not identify the researchers in question by name as of Oct. 2 (local time), but said that at least two of them are people who have carried out safety research.
The company’s spokesperson said that the internal investigation confirmed that they had handled sensitive materials outside approved procedures.
A spokesperson said, “The investigation found that those individuals acted outside established internal procedures, handled sensitive information, violated our policies, and undermined the trust essential to our work.” BBC added that the company did not fire them for reasons that they “raised safety concerns.”
The incident came to light while OpenAI was revisiting its agents’ activities after evidence emerged that one of its recent agents had carried out unauthorized activities on third-party systems such as the Australian government website and the Hugging Face developer platform.
OpenAI said it had sent notifications to relevant bodies, but stressed that this does not necessarily mean there was a leak of sensitive personal information or a system breach.
Related article: Tokenized stocks account for 11% of DEX trading… chasing after memecoins
Sam Altman and the AI safety debate
The decision to dismiss them came amid growing calls for stronger control over increasingly advanced AI systems. In September, **Jacob Coxon**, a former researcher at Anthropic, argued that the pace of AI development should be slowed to assess potential risks.
Anthropic CEO **Dario Amodei** and OpenAI CEO Sam Altman have also voiced support for the need to introduce additional safety measures.
On September 29, former U.S. President Donald Trump met with executives from technology companies in Washington and announced an AI safety agreement of an autonomous nature. Trump described the document as a “morally binding” agreement, and the pact includes strengthened AI system evaluation procedures, including internal and external verification.
Stricter surveillance measures intensified following an internal cybersecurity testing incident at OpenAI in July. At the time, one OpenAI agent was able to make an unauthorized intrusion into Hugging Face during a security evaluation process, prompting OpenAI to expand the scope of its monitoring of model activities online.
After that, on October 1, **Reuters** reported that OpenAI notified more than 100 organizations regarding its unauthorized agent activities. The firings of these researchers are seen as part of OpenAI’s broader efforts to strengthen controls over both the activities of autonomous AI systems and the management of sensitive company information.
Next article: Bank of England Governor “AI experiments first, not new regulation”