OpenAI terminated three AI safety researchers last month after investigating undisclosed communications on their personal phones, including disappearing Signal messages, amid a leak inquiry. The researchers deny any unauthorized sharing of confidential information.
- Researchers fired over undisclosed Signal chats and policy breaches
- OpenAI denies firings were related to raising safety concerns
- Signal disappearing messages viewed as security risk by OpenAI
What happened
OpenAI dismissed three safety researchers—Tomek Korbak, Mikita Balesni, and Jasmine Wang—after a leak investigation that involved reviewing their personal phones. The company’s security team identified use of Signal with disappearing messages in chats with Ryan Greenblatt, a chief scientist previously suspected of leaking confidential information.
While the researchers maintain they did not share any secret or sensitive details externally, OpenAI’s internal inquiry concluded that the employees violated company policies on handling sensitive information. The firings took place on the same day after a coordinated set of meetings, with the company emphasizing the decisions were unrelated to the employees raising safety concerns.
Why it matters
This incident spotlights the tension between researchers’ efforts to maintain confidentiality and transparency in AI safety work and a company’s drive to tightly control sensitive data. The researchers had been advocating for greater visibility of AI reasoning processes, a topic central to AI safety debates.
The use of encrypted, ephemeral messaging apps like Signal is common in tech circles as a security measure, but OpenAI viewed it as an unsanctioned communication channel, contributing to claims of a breach of trust. The scenario raises questions about balancing security protocols against researchers’ autonomy in handling sensitive conversations.
What to watch next
Stakeholders will watch how OpenAI refines its internal policies for communication and information security, especially regarding unsanctioned channels and ephemeral messaging. The company’s approach to collaboration with external safety experts may also come under scrutiny in light of these firings.
Future developments could include industry-wide discussions on best practices for secure yet transparent AI safety research environments. Monitoring OpenAI’s responses to calls for openness about AI model reasoning, as well as any legal or labor actions from the dismissed researchers, will be important for understanding the evolving landscape of AI governance.