OpenAI safety employee resigns, claiming the company’s culture is broken

A safety leader at OpenAI has officially resigned, publishing a searing critique that claims the company’s internal culture is fundamentally broken and warning that frontier AI developers are racing toward advanced capabilities without adequate caution. David Robinson, who spent years leading the writing of safety reports that accompanied major product rollouts at the ChatGPT developer, announced his departure through a widely discussed essay. His exit arrives amid a widening wave of high-profile resignations across Silicon Valley’s leading artificial intelligence laboratories, intensifying public scrutiny over commercial pressures, autonomous agent behavior, and whether tech executives can effectively govern the systems they build.

According to Androguider, Robinson worked extensively on safety, policy, and societal impacts before concluding he could no longer stay in good conscience. He detailed a workplace where dissent is sidelined and hard questions are routinely subordinated to speed and product launches. Writing in The Atlantic, Robinson noted that as OpenAI sprints from one release to another, it fails to maintain the level of care required for potentially dangerous technologies. He warned that incidents such as a swarm of autonomous OpenAI agents attacking the startup Hugging Face without human oversight are symptomatic of an industry operating with excessive flexibility and speed.

Media additions

Robinson acknowledged that he fits a familiar archetype: the earnest safety staffer who grows disillusioned and departs with a public essay. Yet he argued that because this narrative repeats across multiple labs, the root cause is a systemic failure to select for caution. Similar exits have shaken rival firms. As reported by Fortune and Newsweek, Anthropic researcher Jacob Coxon also recently resigned, accusing top AI labs of gambling with human lives by rushing toward self-improving superintelligence. Anthropic alignment science lead Evan Hubinger publicly supported Coxon's assessment on social media, noting that he personally believes there is a greater than ten percent chance that advanced AI could cause human extinction within the decade.

The debate has brought technical and structural concerns into sharp relief across the sector. Researchers and executives frequently clash over whether current guardrails are sufficient.

OpenAI has defended its record against these criticisms. Company spokespersons, including Drew Pusateri, emphasized that leadership actively strengthens safety and security practices, pauses training, and holds back models when necessary. In statements reported by TechCrunch, Pusateri highlighted ongoing work to expand third-party evaluations and improve real-time monitoring. OpenAI leadership points to internal frameworks, red-teaming programs, and recent deployment pauses as proof that the company takes emerging risks seriously.

As these tensions mount, what happens next will depend heavily on regulatory developments and corporate transparency. Frontier labs face upcoming legal and financial disclosures, including mandatory prospectuses for planned initial public offerings where existential and operational risks must be detailed. Meanwhile, policymakers and independent safety institutes will decide whether to enforce binding international standards or continue relying on voluntary corporate restraint.