Quick Answer
- David Robinson, who worked on safety and transparency at OpenAI, resigned. Reports are dated October 3, 2026.
- He published an essay in The Atlantic arguing that OpenAI’s trial-and-error approach to releasing AI is risky as models get more capable.
- OpenAI says it is strengthening security in research and testing, expanding independent evaluator partnerships and upgrading real-time monitoring.
- Many details come from news summaries of his essay, so read the original piece for his exact words.
Another safety voice has left OpenAI, and this time he explained why in public. Here is who Robinson is, what he said and what OpenAI replied.
Who is David Robinson?
Robinson was part of OpenAI’s Safety Systems team. According to Benzinga, he worked on safety and transparency, including system cards, the documents that explain what a model can and cannot do when it launches. He had earlier led the company’s policy planning work. His resignation was confirmed to Business Insider on October 3, 2026.
What did he say?
Robinson published his critique in The Atlantic on October 3, 2026. Summaries of the essay say he argues that:
- OpenAI has succeeded through trial and error, and that approach becomes more dangerous as models grow more powerful.
- The company is not applying the level of care that he believes is needed as it moves quickly from one product launch to the next.
- Safety should not depend on people fixing problems after something goes wrong.
- AI companies should borrow safety practices from fields like nuclear power and aviation.
These points come from secondary reports of his essay, so the wording here is a paraphrase and not a direct quote.
What did OpenAI say?
According to Briefs.co, OpenAI said it is:
- Strengthening security in research and testing.
- Expanding partnerships with independent third-party evaluators.
- Upgrading real-time monitoring.
An OpenAI spokesperson also said the company is working to keep model capabilities within safely manageable limits.
Why this lands now
Benzinga links his exit to growing scrutiny of OpenAI after incidents involving its AI agents, including activity affecting more than 100 organizations. We covered one such case in our report on OpenAI agents reaching government websites. Earlier this month OpenAI also paused training, and in September it cancelled the GPT-6.1 Astra release over safety concerns.
Robinson follows OpenAI’s safety systems head, Johannes Heidecke, who left earlier this year. Benzinga also lists other senior exits in 2026, including the robotics head in March, the applications president in July and the COO in August.
Is this part of a pattern?
Departures from AI labs often come with public warnings. A similar case was Anthropic’s Jacob Coxon resignation. Every case has its own reasons, so it is better to read each person’s statement than to assume the same story.
What we do not know
- OpenAI has not named a successor for the transparency role, according to reports.
- OpenAI has not responded line by line to Robinson’s claims.
- Some details in secondary summaries, such as his exact tenure, could not be checked against the original essay.
FAQ
Who is David Robinson?
A member of OpenAI’s Safety Systems team who worked on safety and transparency, including model system cards.
When did he leave OpenAI?
Reports are dated October 3, 2026, when his resignation was confirmed and his essay was published.
Where did he explain his reasons?
In an essay in The Atlantic.
What did OpenAI respond?
It said it is strengthening security in research and testing, adding independent evaluators and upgrading real-time monitoring.
Is OpenAI still doing safety work?
Yes. OpenAI says it continues to work on safety, and it has other safety leaders. Critics, including Robinson, argue the pace of releases does not leave enough care.
What is a system card?
A public document that explains a model’s capabilities, limits and risks when it is released.
