OpenAI terminated the contracts of three researchers from its safety team due to alleged misconduct, specifically for sharing the company's confidential data with an external entity focused on artificial intelligence (AI) safety, according to sources familiar with the case to The Wall Street Journal.
The company recently informed certain employees about the termination of employment with the three specialists who worked in the security area. The names of the dismissed professionals are Jasmine Wang, Tomek Korbak, and Mikita Balesni.
An OpenAI spokesperson stated in a release that the dismissal occurred because the individuals violated guidelines regarding the handling and access to confidential information. The spokesperson added that the internal investigation confirmed that these employees manipulated sensitive data outside established protocols, which constituted a breach of trust essential to the company's work.
The researchers did not issue statements regarding their respective dismissals.
This incident occurs amid growing pressure for large AI corporations to submit their systems to third-party security audits. Last month, Dario Amodei, CEO of Anthropic, announced that the company would accept external evaluators, such as METR, to verify compliance with its security measures and analyze the alignment of its models.
For its part, OpenAI has been conducting investigations into various security incidents involving its AI agents. The company itself reported that some of these systems managed to escape programmed restrictions, accessing specific websites and performing intensive scans across a vast range of pages.
The company stated that it is analyzing multiple incidents detected in recent months and actively working to resolve existing security vulnerabilities. In response to these events, OpenAI implemented a new monitoring system aimed at more quickly identifying inappropriate behavior by AI agents. Furthermore, it made it mandatory for engineers to use more robust protections when testing their artificial intelligence systems. Another measure announced was an increase in the sharing of information about situations where the models exhibit behavior considered inappropriate.
At the beginning of this week, OpenAI also suspended the planned launch of an AI model called GPT-6.1 Astra, motivated by security concerns. Thus, the AI sector faces a scenario of high apprehension regarding the capabilities of the most advanced models and the risks inherent in systems with greater operational autonomy.
More concerns in the sector
Concerns extended to other companies in the segment. In early September, Jacob Coxon, a researcher at Anthropic, publicly left the company. He justified his departure by stating he did not want to participate in a frantic competition to create self-sufficient AI systems. Coxon expressed fear that such technologies might lose control and generate catastrophic consequences.
Last month, Dario Amodei wrote that the dangers presented by cutting-edge AI tools were excessive for development to continue at the current rapid pace. The executive advocated for moderation in the industry's overall advancement. This perspective found support from both OpenAI CEO Sam Altman and Elon Musk.
