Fired OpenAI researchers deny leaks, point to AI transparency disputes
Three former OpenAI safety researchers have denied allegations of leaking confidential information, suggesting their dismissals were motivated by their advocacy for AI model transparency.
Tomek Korbak, Mikita Balesni, and Jasmine Wang were terminated by OpenAI following an internal investigation that claimed they mishandled sensitive data. The researchers contest these findings, noting that the company failed to provide written evidence detailing the alleged breaches. Instead, they believe their departure is linked to their internal advocacy for greater transparency regarding how AI models process information.
The conflict highlights a growing industry debate over whether external researchers should be permitted to review the internal 'chain-of-thought' processes of AI models. The researchers have argued for the preservation of these audit trails, while OpenAI has moved toward models that store fewer intermediary steps. The company maintains that the terminations were unrelated to the researchers' stated safety concerns.
This incident has raised concerns regarding the future of independent safety audits in the AI sector. The researchers argue that such actions discourage staff from reporting vulnerabilities. Meanwhile, OpenAI states that it has remediated infrastructure security flaws and is preparing to engage in new external collaborations to address safety oversight.