FULL STORY

OpenAI Safety Veteran's Exit Sparks Culture Debate

OpenAI safety veteran David Robinson resigned and published a first-person essay criticizing the company's broken culture. The fallout sparked debate on X, with OpenAI researchers pushing back on misreadings of the team's commitment to safety.

2026-10-04 ~ 2026-10-06 · 2 episodes · 19 posts

Episode 1 · OpenAI Safety Veteran David Robinson Resigns, Saying the Company's Culture Is Broken (2026-10-04, 17 posts)

David Robinson, who was in charge of writing safety reports for OpenAI's product releases, has announced his departure and published an article in The Atlantic titled "I Left OpenAI Because Its Culture Is Broken." After three and a half years at OpenAI, he warned that "the era of trial and error is over," argued AI companies are being "far from careful enough" in developing the technology, and urged the industry not to focus solely on new rules or laws. Reuters, The Guardian, and other outlets covered the story.

Confirmed

  • The departing employee is David Robinson, responsible for OpenAI's product launch safety reports, with 3.5 years of tenure
  • He wrote the piece in The Atlantic in the first person, stating outright that the company culture is "broken"
  • This is the latest case in the ongoing talent exodus from OpenAI's safety team

Why it matters

  • Ben Todd quoted the employee as noting that at OpenAI, founded roughly 11 years ago, 3.5 years of tenure ranks among the longest, calling the attrition "brutal"—reflecting the constant turnover in core teams at frontier AI labs
  • Former policy researcher Miles Brundage confirmed the attrition is indeed brutal, but noted that given the company's exponential growth, such a claim would hold in almost any case; Ben Todd added the caveat that the statement holds only if it refers to "the top 5% by tenure," and chart-based presentations can easily exaggerate how rare it is. By Ben Todd's estimate, Anthropic has over 200 employees with more than 3 years of tenure, suggesting high turnover is not unique to OpenAI
  • The episode echoes earlier criticism of OpenAI's "iterative deployment" trial-and-error culture, highlighting the tension between safety voices and the pace of commercialization

Episode 2 · OpenAI Safety Culture: Everyone's Responsibility, Not Nobody's (2026-10-04, 2 posts)

OpenAI researcher csvoss pushed back on misreadings of her safety stance, clarifying that safety culture means everyone's responsibility, while former colleagues discussed how team culture must keep pace with rising safety standards.