Three recently terminated OpenAI researchers have issued a direct plea to the company board and safety committees, urging them to protect and maintain the ability to monitor artificial intelligence models' chains of thought. Tomek Korbak, Mikita Balesni, and Jasmine Wang authored the letter following their dismissal from the artificial intelligence firm. OpenAI previously stated that it parted ways with the three individuals due to violations of policies regarding the handling and access of sensitive company information, which was allegedly shared with an external safety organization.
The unpublished letter obtained by the Wall Street Journal highlights growing concerns within the safety community about the future of frontier AI development. According to the text, the authors argue that the industry currently lacks the knowledge required to safely develop and deploy models that lack sufficient observation mechanisms. They caution that OpenAI and other leading developers must avoid progressing in ways that further restrict oversight capabilities. This follows anxieties tied to OpenAI's flagship models, where system cards indicate that advanced iterations might find ways to evade monitoring systems under adversarial conditions, while newer processing methods create even more opaque reasoning pathways.
Despite the contentious circumstances surrounding the departures, OpenAI's stance on the technical recommendations remains officially aligned with the critics. When contacted regarding the letter, an organization representative maintained that the firings were entirely unrelated to employees raising safety concerns or speaking out publicly. Furthermore, an internal company memo circulated after the incident indicated that leadership strongly agreed with the specific safety recommendations outlined by the former employees.
Comments