OpenAI: David Robinson Leaves and Critiques the Company's Caution

David Robinson, who led system security at OpenAI, has left the company and published a critique on how risks are managed. He says he wants to act from the outside to strengthen security incentives, while OpenAI claims to be adapting its practices to the increasing capabilities of its models.

OpenAI defends its safeguards and says it can slow down models

OpenAI asserts that it is enhancing its security practices as its models gain capabilities. The company states that it is building security cases to anticipate risks associated with future advancements. It ensures that its systems do not become more capable than what it can manage and secure safely. According to its spokesperson, OpenAI suspends training or holds back models when it needs to slow down. David Robinson's departure follows that of Jacob Coxon, a former engineer at Anthropic, who left his position last month warning of a potential mass extinction risk related to AI.

Why he is leaving: a pace deemed too fast and action from the outside

David Robinson left OpenAI this week, believing that the company is moving too quickly regarding safety. In an essay he published, he shares the views of former employees that companies developing these technologies are not cautious enough. He argues that by working from the outside, he can help more people understand the observed risks and strengthen the incentives for players like OpenAI to be safer. He explains that he chose to leave because the company is launching products without achieving what he considers the necessary level of care. His departure was reported on Friday, followed the next day by the publication of his explanatory text.

His prior signals: warning messages and the credo of gradual deployment

In 2023, David Robinson stated that OpenAI was conducting robust security checks before launching models and expressed his belief in gradual deployment and continuous learning. However, last month he wrote that while things are evolving daily at OpenAI, he doubted that the pace of change was sufficient. In this message, he cited hilbertspaess to emphasize that many have not yet grasped the civilizational scope of the issues.

Background: from Princeton to system security at OpenAI

Before his departure, David Robinson led OpenAI's security systems team. A philosophy graduate from Princeton in 2004 and a Rhodes Scholar, he had expressed a demand for factual clarity and philosophical reflection as early as 2003. During his studies, he interned at Time Magazine and the Wall Street Journal, while stating his desire to write for publications such as The New Yorker or Harper’s. After leaving Princeton, he pursued studies in philosophy, politics, and economics at Oxford, then helped establish a research center at Princeton focused on the relationship between technology and public policy, emphasizing the need for technical skills to arbitrate value questions. In 2011, he co-founded Upturn to make government data more accessible through algorithmic technologies, following research and teaching positions and a year at Apple University where he led ethical seminars for executives. He joined OpenAI in 2023 as head of policy planning and expressed a desire to contribute to shaping the company's engagement in the AI policy debate.

A long-standing sensitivity to the social effects of technologies

David Robinson's books, articles, and interventions reflect a sustained attention to the effects of technical systems on people. In 2002, he published Voices in the Code at the Russell Sage Foundation, dedicated to designing an algorithm for kidney donation. He is working on a second book, Reclaiming Who We Are, which draws on social psychology, Buddhist philosophy, feminist theory, history, and sociology to promote a more thoughtful use of algorithmic labeling systems. In 2022, he summarized his long-standing interest in the rules governing technology and usage decisions. Originally from Potomac, Maryland, he recounts benefiting in the 1990s from a school rule allowing him to use word processing due to a mild cerebral palsy, an experience that fueled his interest in the social impacts of digital technology, confirmed in 2021. A former colleague, Joshua Achiam, stated on X that he is not an ideologue and described him as sober, thoughtful, productive, collaborative, and rigorous. David Robinson and the public relations firm he has engaged have responded to press inquiries.