Artificial Intelligence
OpenAI safety leader David Robinson leaves company
David Robinson left OpenAI and later argued in an essay for The Atlantic that the company and other AI developers were not being careful enough with powerful models.
- David Robinson left OpenAI last week and later published an essay in The Atlantic.
- He had worked on safety transparency, including system cards for OpenAI's models.
- In the essay, he said companies building AI were not being careful enough and that OpenAI was moving too quickly.
- OpenAI said it was expanding outside evaluation and live monitoring, while Sam Altman said safety and alignment had to stay ahead of capability.
- The company also said it had ended its relationship with three researchers for sensitive-information policy breaches, and Johannes Heidecke had left earlier in the year.
Robinson left OpenAI last week and later published an essay in The Atlantic. He said companies building AI were not being careful enough and that the culture around the technology encouraged overconfidence.
He had worked on safety transparency and on system cards for OpenAI's models. He said he did not take the decision lightly and that he still believed AI could be useful.
OpenAI said it was trying to keep its models within what it could safely manage, and that it paused training or held back models when needed. It said it had expanded work with outside evaluators and was improving live monitoring for troubling behaviour.
The company also said on Thursday that it had ended its relationship with three researchers for breaching policy on sensitive information, and Johannes Heidecke, its safety head, had already left earlier in the year. Sam Altman said at the company's developer conference on Tuesday that safety and alignment had to stay ahead of model capability.
Named in this story
People
- David Robinson
- Former OpenAI safety systems leader who quit and then criticised the company
- Sam Altman
- OpenAI chief executive who said safety had to stay ahead of capability
- Johannes Heidecke
- OpenAI safety head who left earlier in the year
Companies
- OpenAI
- The company Robinson left and later challenged over its safety culture
Organisations
- The Atlantic
- Published Robinson's essay
How the source tells it
The piece was written in a critical, urgent register, with Robinson's warnings and OpenAI's response setting up a dispute over caution and control.
- alarm worst-case language about dangerous systems and models that might not do what their makers want
- urgency pace-of-change wording built around sprinting, slowing down and staying ahead
- reassurance a corporate defence promising controls, pauses and earlier detection of concerning behaviour
- moral framing the issue was cast as a test of humility and responsibility