Former White House AI and Crypto Czar David Sacks said the White House grew concerned that Anthropic’s AI systems could pose cyber weapon risks after reported jailbreak issues and escalating disputes over the rollout of its "Fable" model.

White House Alarm Over AI Jailbreak Claims

On Saturday, speaking on the "All-In Podcast," Sacks said he received readouts from White House officials and private companies suggesting rising national security concerns that Anthropic’s models could function as a "cyber weapon."

"What this comes down to is a couple of things," Sacks said, describing internal government concern that the company’s models had advanced cyber capabilities.

He said officials were especially concerned after Anthropic CEO Dario Amodei reportedly described an earlier system as having powerful offensive cyber potential.

Sacks said this framing contributed to officials viewing the technology as potentially dual-use.

According to Sacks, tensions escalated when testing partners allegedly identified jailbreak vulnerabilities in "Fable," a later model with safety guardrails.

"It wasn’t the White House who came to the conclusion… it was private companies who were testing Fable," he said, adding that concerns were then escalated to federal officials.

Sacks also claimed there were attempts by the White House to contact Anthropic to address the issue, saying discussions became strained over how serious the vulnerability was.

"They tried to call Anthropic… and the TL;DR is Dario refused," he said, characterizing the breakdown as a trust issue. Anthropic has not publicly confirmed those claims.

US Split Over AI Policy And Anthropic Ban

Earlier, Sriram Krishnan stepped down as White House senior AI policy advisor at the end of June after helping shape the administration’s AI strategy focused on expanding data center infrastructure.

He credited President Donald Trump’s leadership and Sacks and planned to remain involved in AI policy through an outside institution.

Meanwhile, the U.S. government was divided over Anthropic’s AI systems.

The Pentagon blacklisted the company from military use over national security concerns, labeling it a "supply chain risk," while Trump signaled openness to a potential deal.

To address broader concerns, White House officials drafted a national security AI policy requiring agencies to use multiple AI providers and limiting contractor influence over military command structures.

The administration also considered ways to bypass Anthropic’s restrictions to allow potential use of its AI model, Mythos.

Photo courtesy: Shutterstock

Disclaimer: This content was partially produced with the help of AI tools and was reviewed and published by Benzinga editors.