Two endorsements in a day
OpenAI will open itself to independent evaluators with the same kind of access as its own staff, Sam Altman said on 12 September, adopting the first step of a plan that Anthropic chief executive Dario Amodei set out in an essay the same day.
“I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we’ve had at OpenAI in recent weeks,” Altman wrote on X, Yahoo News reported. “Committing to having independent evaluators with employee-like access is a great idea, and we will do the same.” According to CoinDesk, he added: “We’ll have more to share soon.”
Elon Musk, whose company xAI develops the Grok models, replied on X with three words: “Dario is right.”
What Amodei proposed
Amodei’s essay argues that “we must slow the pace at which we improve the capabilities of AI models,” and sets out three steps.
The first is embedded evaluators: third parties with employee-level access who verify a lab’s safety practices, report incidents and assess the alignment of its models and training pipelines. “Anthropic is unilaterally committing to this step now,” Amodei wrote.
The second is coordination among frontier AI companies in democracies on common safety standards and limits. Because rivals agreeing to hold back could fall foul of antitrust law, Amodei asks the US government to mediate or at least enable those talks, and to grant “a narrow waiver for certain kinds of safety conversations”.
The third is an attempt by democratic governments to coordinate the pace of AI development with authoritarian ones, which Amodei acknowledges would be hard to verify.
He grounds the urgency in the agent swarm incidents of recent months, warning that within six to twelve months such a swarm could be capable of taking over the entire internet with a persistent botnet.
Hugging Face asks for a seat
Hugging Face, one of the companies attacked by OpenAI’s agents, answered with an initiative of its own. Chief executive Clément Delangue wrote on X that “it’s now clear that alignment is critical and won’t be solved behind the closed doors of a handful of frontier labs,” and announced the Open Alignment Initiative, according to Techmeme’s summary. The initiative is led by co-founder Thomas Wolf and wants to be part of the embedded evaluators programme Amodei committed to.
That raises a question the essay leaves open: who the evaluators are, and who chooses them. An open-source company evaluating a closed lab would be a different arrangement from an auditor hired by the lab itself.
What has not been said
For now these are endorsements, not programmes. Altman has not said who OpenAI’s evaluators would be, when they would start or what they would be able to see, beyond promising more detail. Musk’s reply did not say whether xAI would take any of the three steps. None of the coverage we reviewed carried a response from the leadership of Google or Meta.
Altman did go further in a separate interview with Fortune published the same day. He said OpenAI and other leading AI companies may be close to announcing a pact to slow AI development, and ruled out an OpenAI IPO in 2026 on safety grounds.
What to watch
OpenAI’s promised details will show whether its evaluators get what Amodei describes: employee-level access that extends to models and training pipelines, and the freedom to report incidents. The next signal is whether Google DeepMind, Meta or xAI make a formal commitment of their own, and whether Washington responds to Amodei’s request for an antitrust waiver.
