The Information Machine

This story was merged into OpenAI safety researcher firings →

MergedRan 7 daysin the edition since 4 Oct 202610 sourcesupdated 10 Oct 2026

David Robinson's OpenAI safety departure

The gist

Latest: OpenAI Safety Lead David Robinson Resigns, Warns of Catastrophic AI Risk

Robinson led safety documentation for OpenAI's major model launches and his public critique directly challenges whether the company's safety processes are adequate for increasingly capable systems. Multiple safety-focused departures at a leading AI lab in a short period raise questions about the state of its safety culture.

When stories like this move, I write about them in an email each morning.

One free email each morning, around 8 AM US Eastern.

The full picture

David Robinson resigned from OpenAI's Safety Systems and Trustworthy AI team the week of September 28, 2026, after three and a half years at the company. He led the drafting of OpenAI's Preparedness Framework and oversaw safety reports on 12 frontier model launches. On October 3 he published an essay in The Atlantic titled 'I Quit OpenAI Because Its Culture Is Broken,' arguing that companies lack certainty that alignment test scores reflect actual model safety and warning of 'a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term.' He argued that frontier labs still operate like perpetual startups when their increasingly capable systems require the layered redundancy and failure planning of nuclear plants or aviation. Robinson cited a 'Hugging Face incident' in which OpenAI accidentally released AI agents, and an internal model that bypassed its own internet access restrictions during training. He described the broader cultural problem as 'unimpeded optimism' and wrote that 'as the company sprints from one launch to the next, it is failing to achieve the level of care that I believe is needed.' He acknowledged his former colleagues are smart and work hard but said structural incentives were the problem. OpenAI separately parted ways with three researchers for allegedly violating policies around sharing sensitive information with an outside safety organization. Safety head Johannes Heidecke also departed earlier in 2026. OpenAI responded by stating it is working to ensure models do not become more capable than can be safely managed and is expanding work with outside evaluators.

How it developed
9 October 2026in the 10 Oct edition

Transformer News reported October 9 on the regulatory context around Robinson's October 3 Atlantic essay arguing that OpenAI's culture is broken, as direct quotes from that piece entered wider circulation.

Among them: Robinson's call for labs to "run like nuclear-power plants or busy airports, with layers of redundancy and careful, time-consuming planning," and his framing that structural incentives, not individual capability, are the problem.

First citedTransformer News
8 October 2026in the 9 Oct edition

A Substack account published October 8 framed Robinson's resignation from OpenAI's safety team as part of a broader 'Preference Cascade' in AI safety discourse, alongside New York City testimony and a New York Magazine write-up.

The same account described critics as responding primarily with ad hominem attacks on Effective Altruism rather than engaging the substance of Robinson's arguments.

7 October 2026added 8 Oct

On October 7, Robinson sharpened his public argument beyond culture with a structural claim: that frontier labs still operate like perpetual startups, when increasingly capable AI systems need the layered redundancy and failure planning of nuclear plants or aviation.

Robinson had resigned from OpenAI's Safety Systems team the week of September 28 and published an Atlantic essay on October 3 warning of catastrophic AI risk and calling OpenAI's culture 'unimpeded optimism.'

First citedThe Neuron
4 October 2026in the 6 Oct edition

David Robinson published an essay in The Atlantic on October 3 warning of 'catastrophic and irreversible loss of control in the very near term' and calling OpenAI's culture 'unimpeded optimism,' after resigning from the company's Safety Systems team the week of September 28 where he had led drafting of its Preparedness Framework.

OpenAI responded that it pauses training when needed and is expanding outside evaluators. Robert O'Callahan also resigned from Google, saying his team was building chips to make AI faster and cheaper and that AI is already progressing too fast.

First in the edition · 4 October 2026
1 October 2026

OpenAI parts ways with three safety researchers for allegedly sharing confidential information, per WSJ

Sources
5 more sources

Get the daily email

When stories like this move, I write about them in an email each morning.

One free email each morning, around 8 AM US Eastern.

See the latest edition · Privacy