David Robinson, a former safety official who authored the risk assessment reports accompanying major model releases at OpenAI, has resigned from his position, warning that the artificial intelligence industry's internal culture is fundamentally broken. Writing in an editorial for The Atlantic, Robinson argues that the tech sector's high-speed development ethos prioritizes rapid expansion over systemic risk management, leaving critical vulnerabilities unaddressed.
According to Robinson, Silicon Valley frontier labs operate with "extreme confidence" and "perpetual sprints," pursuing ever-larger models with an optimism that routinely underestimates potential harms. He asserts that minor regulatory additions will be insufficient to address these deep-seated issues. Instead, he urges AI companies to cultivate humility and adopt safeguards akin to high-consequence industries, arguing that frontier laboratories should be run like nuclear power plants or busy airports—built with layers of redundancy and careful planning so that inevitable human errors do not lead to disaster.
Robinson’s public departure reflects a broader wave of unease among safety researchers at major AI developers. He is the latest in a growing list of safety staff who have recently left high-profile firms, including Jacob Coxon and Joe Benton from Anthropic, as well as Robert O’Callahan, Bilal Chughtai, and Josh Engels from Google DeepMind. Their resignations emphasize a rising tension between internal safety teams and tech companies pushing to advance artificial intelligence models at a rapid pace.