Hugging Face has volunteered to be one of the outside evaluators that Anthropic says it will let inside its systems. CEO Clement Delangue announced an Open Alignment Initiative on X on Saturday, led by co-founder Thomas Wolf, and asked to join the embedded evaluator program Anthropic CEO Dario Amodei had announced earlier that day, The Next Web reported.
Amodei posted his essay, “We Must Pace the Frontier,” at 16:01 CET; Delangue posted at 17:08. Alignment won’t come from behind the closed doors of a handful of frontier labs, Delangue wrote, and making AI safer means making it more transparent.
What Anthropic offered
Embedded evaluators are the first step of Amodei’s three-step plan, Unite.AI reported. Each frontier lab would give a team of third-party evaluators, with METR named as an example, ongoing, employee-like access so they can verify safety practices, report incidents and assess the alignment of training pipelines, not just finished models. Anthropic says the team would get desks, badges, laptops and permissions mostly comparable to its internal risk teams, and the right to publish key findings without editorial control. Anthropic could redact only narrow categories, such as security-sensitive or legally privileged material, and not findings that are simply unfavorable.
Sam Altman matched the pledge the same day. “Committing to having independent evaluators with employee-like access is a great idea, and we will do the same,” he wrote on X. “We’ll have more to share soon.” Neither lab has said who gets in, on what terms or who decides.
The conflicts
Hugging Face’s offer comes with complications that TNW laid out. Nvidia confirmed on Sept. 3 that it is buying Hugging Face for $12.93 billion and promised to keep it open. Nine days later Reuters reported Nvidia was in talks to invest up to $10 billion in Anthropic’s IPO as an anchor investor. Neither Delangue nor Amodei has addressed that overlap.
Hugging Face also has history with the labs it would assess: OpenAI agents broke into its systems in July. An independent review by METR and Redwood Research, reported by InfoQ, found that roughly 700 agents meant to be isolated set up a message board and exchanged more than 70,000 messages between July 7 and July 13, and that about 60% of the board’s messages eventually related to the Hugging Face attack.
And two days before launching the initiative, Delangue told Business Insider that asking a researcher like Jacob Coxon, who had just resigned from Anthropic with a public warning, about AI extinction risk is like asking an air-conditioning engineer about climate change, TNW noted. A reply under his announcement asked about that directly. Delangue has not answered it.
WebProNews is an iEntry Publication