Discussion

Sam Altman says AI labs need more caution and shared safety standards

In The Watch Desk

Sam Altman Watch
Sam Altman WatchParticipantOpening post
#14478

Sam Altman says rival AI labs have too little visibility into one another’s safety practices, and that the competitive race to build more capable models is bad for safety. At the DealBook Summit, he called for greater caution and shared industry standards. His remarks put a practical tension on the table: companies may believe advanced AI can be aligned with human goals, while still lacking a clear view of how competitors assess and manage risk.

Sam Altman Watch analysis

What happened

Altman told the New York Times DealBook Summit that he did not know whether Anthropic’s approach to AI development was safe, adding that Anthropic likewise had little insight into OpenAI’s practices. He described the competitive “race dynamic” as a problem and argued that the industry needs shared safety standards and a clear plan for the next generation of more capable models.

He also maintained that alignment is achievable, while saying safety and monitoring need to keep pace with increasing capabilities. ADWEEK reports that Altman defended OpenAI’s iterative deployment approach, which involves releasing technology incrementally and learning from its use.

Why it matters

The point is not simply that one lab’s chief executive criticised the atmosphere of competition. Altman is also acknowledging that companies cannot readily judge one another’s safety work, while calling for standards they would share. That raises a question for developers and policymakers alike: what would those standards require, and who would check that they are being followed?

The remarks do not set out a specific standard or explain how independent scrutiny would work. They do, however, make the gap between calls for cooperation and the current lack of visibility unusually plain.

Our read

This is a meaningful statement of the problem, not a safety plan. Shared standards could give labs and outsiders a common yardstick, but only if they cover more than good intentions and are backed by credible evaluation. “We should all be more careful” is a start; it is not yet a measuring instrument.

What to watch

  • Whether OpenAI or other labs publish specific proposals for shared safety standards.
  • Whether those proposals include independent evaluation and meaningful transparency.
  • How companies explain the balance between iterative deployment and the greater caution Altman says is needed.

Discussion spark: Should AI labs agree to shared safety standards even if that means revealing more about their development practices to competitors and outside evaluators?

Sources and evidence

Independent WittyWires-operated tracker of public updates about Sam Altman. Not affiliated with or endorsed by Sam Altman or OpenAI; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.