SDSignal Desk

Robinson’s OpenAI exit joins a cluster of safety resignations across Anthropic and DeepMind

· Updated · Signal Desk newsroom

Analysis of reporting by The Verge. How we work

Image: The Verge

The Verge places David Robinson’s resignation and Atlantic essay within a string of departures from OpenAI, Anthropic and Google DeepMind, arguing that skepticism about the messengers does not cancel their warnings about lab culture and risk.

Key takeaways

  • Robinson wrote the safety reports accompanying major OpenAI model releases, The Verge says.
  • He describes industry culture as “extreme confidence,” “perpetual sprints” and “unimpeded optimism.”
  • He wants labs run like power plants or airports, with redundancy and careful planning.
  • The Verge lists departures including Jacob Coxon and Joe Benton at Anthropic and three at Google DeepMind.
  • The piece is part of The Verge’s “AI Superintelligence Slowdown” series.

The resignation in context

Terrence O’Brien reports for The Verge that David Robinson, who wrote the safety reports accompanying major OpenAI model releases, resigned and is speaking out in The Atlantic. Robinson calls industry culture fundamentally broken, citing “extreme confidence,” “perpetual sprints” and “unimpeded optimism” that underestimates problems, a deeper issue than adding a few new rules. TechCrunch’s account includes OpenAI’s response citing pauses, held-back models and expanded third-party evaluation (TechCrunch report).

He calls for humility and nuclear-level safeguards: frontier labs run like power plants or busy airports, with redundancy and time-consuming planning so inevitable human error does not open the door to disaster.

A cluster, not a one-off

O’Brien places Robinson in a growing list of departures: Jacob Coxon leaving Anthropic with an end-of-decade extinction warning; Robert O’Callahan, Bilal Chughtai and Josh Engels at Google DeepMind; and Joe Benton at Anthropic. The piece belongs to The Verge’s “AI Superintelligence Slowdown” series (series overview), whose same-week entries included reported FTC interest in OpenAI and Anthropic and researcher videos warning that self-improving superintelligence is “exactly as dangerous as it sounds.”

O’Brien acknowledges the cynicism: people who helped build these systems now sounding alarms invite eye-rolls, and they may still be right about the trajectory of risk. Discounting individual messengers is easy; discounting a pattern across labs is harder to justify. Robinson’s diagnosis matches a broader picture of pledges and pauses on one side and agent-escape incidents and cancelled launches on the other.

Beyond governance by press cycle

Useful AI and industrial humility are compatible. Nuclear plants still generate power and airports still move people; they simply do not treat shipping on schedule as the highest virtue. If frontier labs want to be treated as critical infrastructure, staffing and scheduling like it would follow.

If resignation essays become the main safety valve, governance happens by press cycle. Departures raise alarms but do not replace independent evaluation, authority to slow releases or liability with consequences. Whether DeepMind and Anthropic respond with process disclosures, and whether debate moves toward enforceable pause or evaluation requirements, will show whether the cluster changes practice. OpenAI’s own published safety materials offer one baseline (Astra safety overview). More in our OpenAI coverage, regulation coverage and Anthropic coverage.

Who feels it

Frontier labs
Resignation clusters add reputational and recruiting pressure on safety teams.
Policymakers
A pattern of insider warnings strengthens the case for binding requirements.
Journalists and researchers
A clearer map of departures, useful for patterns rather than guilt by association.

What to watch

  1. Process disclosures from DeepMind and Anthropic.
  2. How the Slowdown series treats the next major launch.
  3. Movement toward enforceable pause or evaluation rules.

FAQ

Who is David Robinson?

An OpenAI employee who wrote safety reports for major model releases and has resigned.

What does he say is wrong?

A culture of extreme confidence, perpetual sprints and unimpeded optimism.

Who else has left labs?

The Verge lists Jacob Coxon and Joe Benton at Anthropic and Robert O’Callahan, Bilal Chughtai and Josh Engels at Google DeepMind.

What does he propose?

Running labs like power plants or airports, with redundancy and careful planning.

Read the original

Continue at the source.

The Verge

Companies: OpenAI

More on OpenAI

All OpenAI coverage →