IA · 9 October 2026 · 4 min read
Open Clash at OpenAI: Fired Safety Researchers Warn of an Internal Culture of Fear
In brief: Tensions inside OpenAI have reached a boiling point following the firing of safety researchers Jasmine Wang, Tomek Korbak, and Mikita Balesni. While leadership maintains the dismissals stem from serious protocol violations regarding sensitive data, the trio published a joint open letter warning that internal retaliation has created a chilling effect among remaining engineers. The dispute highlights the escalating friction between commercial secrecy and safety governance at frontier AI labs.
by Team Mocchi's
OpenAI Doubles Down as Misconduct Claims Face Pushback
The internal balance between scientific safety oversight and corporate control has broken into full view. At the center of the dispute is the dismissal of three prominent AI alignment researchers: Jasmine Wang, Tomek Korbak, and Mikita Balesni. Over recent days, OpenAI formalized the terminations, pointing to alleged violations of company rules concerning access to and sharing of proprietary internal records.
As reported by The Verge, the company has held firm on its stance, asserting that an internal investigation uncovered a «significant breach of trust» that crossed acceptable operational boundaries. OpenAI firmly rejected claims of censorship, maintaining that the action had nothing to do with staff raising concerns over frontier model risks, but was instead an enforcement of strict data-handling policies.
An Open Letter Warns of a Chilling Effect Across Teams
The leadership's explanation was swiftly contested by the three researchers, who addressed an open letter directly to OpenAI's Safety and Security Committee, Safety Advisory Group, and Mission Advisory Council. As detailed by TechCrunch, Wang, Korbak, and Balesni denied mishandling information maliciously, explaining that their interactions with external AI safety specialists aligned with standard practices that were considered normal company behavior until very recently.
In their letter, the scientists argued that frontier AI systems cannot be evaluated safely without open cooperation across specialized external institutions. Their most pressing warning focuses on corporate culture: the group warned that their dismissals have triggered a widespread chilling effect, leaving current staff uncertain of where they stand and reluctant to voice technical disagreements or flag emerging system risks.
Commercial Secrecy Versus Scientific Scrutiny
The confrontation exposes an unresolved tension at the core of the AI boom. Leading labs now function as multi-billion-dollar commercial platforms that guard algorithms, trade secrets, and operational telemetry behind stringent non-disclosure pacts. Conversely, safety research inherently depends on transparency, peer validation, and early discovery of systemic failures before models deploy broadly.
As autonomous agents and recursive capabilities mature, the boundary between proprietary corporate information and vital public safety warnings becomes dangerously narrow. Observers caution that as internal dissent faces increasing bureaucratic hurdles, external trust in industry self-regulation could deteriorate, complicating compliance discussions with international policymakers.
Mocchi's take
For enterprise teams like ours building software architectures atop frontier foundation models, this controversy underscores vendor governance as a direct operational risk. When internal safety divisions lose room to challenge commercial roadmaps independently, relying blindly on vendor self-assessments becomes a questionable strategy. Italian enterprises integrating high-level AI must implement their own verification layers, independent monitoring sandboxes, and fallback frameworks to ensure real-world resiliency.