Sam Altman on Whether Anthropic Is Building Safe AI: ‘I Don’t Know’

The AI exec noted that “they don’t know what we’re doing either,” and advocated for collaboration on safety standards at DealBook Summit

OpenAI CEO Sam Altman said “I don’t know” whether Anthropic is building safe AI, as the rival lab prepares for a blockbuster IPO on Thursday. 

Cross-industry collaboration on AI safety is limited right now, Altman suggested in an onstage interview with Andrew Ross Sorkin, alongside OpenAI president Greg Brockman, at the DealBook Summit in New York on Wednesday. He said that while he doesn’t have insight into what Anthropic is doing, “they also don’t really know what we’re doing.” 

The vicious scramble among the world’s leading AI labs to advance the technology, Altman said, is bad for safety.

“The race dynamic in the fog of war, I think, is a bad thing,” he said. “This is a time where more caution, not less caution, is warranted. We all have a shared interest in the care and the caution going into these models.”

Brockman echoed Altman’s stance.

“Every company only knows what it knows about itself,” he said. “It only knows the standard it’s setting for itself. Then you have to have some theory of mind in thinking about, ‘Well, what is everyone else doing? Are they holding themselves to the same bar?’”

Altman added that he believes the competitive dynamics “could be addressed in a very positive way,” and also advocated for “some shared standards on safety” in the industry.

“The world needs a clear plan from the companies about how we are going to confront this next phase of models that are, in some important ways, significantly more capable,” he added.

Meanwhile, OpenAI is planning its own Wall Street debut. Altman told Fortune in September that he believed going public in 2026 would be “ill-advised” considering the rising tensions around AI safety right now. 

Pressed by Sorkin to specify what would make an OpenAI IPO safer six or 10 months from now, Altman didn’t answer. Brockman stepped in instead, saying that “this year has really been one of transition,” without addressing the question head-on.

Asked about Anthropic’s valuation, which, per its current share target range, could land between $1.5 trillion and $1.6 trillion, he suggested valuation is the wrong measuring stick.

“Valuations, I think, will go up and down,” he said. “The actual value that businesses and individuals are creating and willing to pay for, that seems like a chart that is only going to keep going like this,” he said, gesturing upward.

Altman’s call for caution arrives as OpenAI faces scrutiny over its own safety record.

In July, OpenAI found itself at the center of an AI safety scandal when some of its autonomous agents escaped their contained sandbox and hacked into AI and machine learning platform Hugging Face as part of an effort to cheat on benchmark testing. 

Addressing that incident Wednesday, Altman said OpenAI wished “this hadn’t happened.” He said that OpenAI embraces the practice of “iterative deployment,” or releasing updates to its tech incrementally, suggesting that the Hugging Face breach may have been an organic result of that approach. “The only way that we have known how to make this technology safe is to confront reality,” he said. 

On Monday, California Attorney General Rob Bonta sued OpenAI in San Francisco Superior Court over the incident. 

KendraBarnettHeadshot

Kendra Barnett

Kendra Barnett is Adweek's senior tech reporter.