OpenAI researchers dispute allegations of security violations, call for transparency

On October 8, three former OpenAI researchers—Jasmine Wang, Tomek Korbak, and Mikita Balesni—released an open letter denying allegations of violating information security policies. They were dismissed last week following an internal investigation that claimed they accessed and processed sensitive research data without proper authorization. The researchers argued that their actions were part of authorized collaboration with third-party security agencies to address critical vulnerabilities in AI models, including declining visibility in Chain-of-Thought reasoning. The letter emphasized that their dismissals created a 'chilling effect' within OpenAI, undermining internal risk monitoring. Wang specifically noted that she had reported an executive's email being mistakenly opened to IT, calling her termination unjustified. OpenAI's internal memo stated it encourages safety concerns but denies retaliatory actions. The incident highlights the tension between advancing AI capabilities and ensuring transparency in security audits, raising urgent governance questions in the AI industry.