OpenAI fires three safety researchers over information policy breach

By Splitfeed EditorialLast updated Oct 9, 2026, 12:10 UTC

On October 9, 2026, OpenAI terminated three safety researchers, citing violations of its sensitive information policy and what the company called a significant breach of trust. The researchers disputed those characterizations, stating publicly that their dismissals were connected to warnings they had previously raised about AI safety risks. OpenAI denied that the terminations constituted retaliation for speaking out. Separately, Sanghyun Lee, OpenAI's Asia-Pacific policy head, was reported to be leaving the company after roughly six months in the role, adding to a broader picture of personnel movement at the organization.

The dispute matters beyond the immediate personnel decisions because it touches a structural tension that has followed OpenAI for years: whether internal safety concerns receive genuine protection or whether raising them carries professional risk. If researchers who flag risks can be dismissed on policy grounds in ways they describe as pretextual, others inside the company may calculate that candor is costly. That chilling effect, and whether it is real or overstated, is the central factual question that neither side has resolved publicly. The simultaneous departure of a senior policy figure adds uncertainty about the company's direction in key regulatory markets.

The Anti-AI camp has been the most vocal. Accounts including ainieuwtjes.bsky.social and wmur9.bsky.social framed the firings as evidence of a company that prioritizes corporate interests over safety, while privpros.bsky.social and infosecbriefly.bsky.social both warned that the dismissals will chill internal candor and safe development. The account wabisabi4robots.bsky.social characterized OpenAI's approach to trust and safety as completely dysfunctional. A reference from amistadsociety.bsky.social to Ezra Klein's interview with David Robinson, who reportedly quit over concerns that OpenAI's technology poses risks to humanity, placed the firings within a longer pattern of safety-focused departures. The Pro-AI camp, represented here by OpenAI's own account, emphasized continued collaboration with independent safety organizations and external assessors, framing the company's safety infrastructure as intact. The Middle Ground, represented by aidailypost.com, held the question open, noting the debate around data handling and transparency without reaching a verdict.

The argument will be shaped by what follows. Any formal whistleblower complaint, regulatory inquiry, or public account from the three researchers detailing the specific warnings they raised would provide the clearest test of whether OpenAI's stated rationale holds. How the company handles internal safety processes going forward, and whether further departures occur, will also serve as signals that observers on all sides are likely to watch closely.