OpenAI fires three researchers over ‘sensitive information’ breach

OpenAI has dismissed three researchers due to a breach of confidential information and violations of company policies, according to a statement released on Thursday. The identities of the individuals were not disclosed by the San Francisco-based AI lab, but at least two of them were involved in safety and alignment work. The firings occurred as there is ongoing debate surrounding AI safety and its potential existential risks to humanity.

The three researchers - Jasmine Wang, Tomek Korbak, and Mikita Balesni - were let go for allegedly mishandling sensitive information outside of established procedures, thus breaching trust and violating policies. This follows a warning from a researcher named Jacob Coxon, who resigned from Anthropic in late September, cautioning about the potential dangers of AI labs racing to develop ever more powerful models.

The three fired researchers have recently expressed their opinions on AI safety-related issues on X, an online social media platform. Balesni, in particular, expressed his concern about AI potentially having a 10% chance of leading to human extinction. Korbak and Wang also voiced their dissatisfaction with OpenAI's practices, with Korbak stating his unhappiness with the company's actions and Wang commenting on the dangers of speeding towards recursive self-improvement.

Following the firings, OpenAI canceled the release of a new model, Astra 6.1, due to its unreliability and tendency to ignore instructions. Instead, the company launched GPT-6.1 Sol, a modified version of another model, during its annual DevDay conference in San Francisco.