Anthropic Asked for Regulation. Washington Went Much Further

Anthropic, a leading artificial intelligence firm at the forefront of AI safety advocacy, is facing a critical juncture as its proactive stance on regulation inadvertently triggers government intervention. The company, now valued at nearly $1 trillion, has consistently championed robust oversight for advanced AI models. However, for the second time this year, Anthropic finds itself under scrutiny from the current administration, this time regarding the potential national security implications of its latest AI systems.

Late Friday, just hours after SpaceX concluded its highly anticipated initial public offering, Anthropic announced it had received an export control directive. This directive mandates the company to immediately suspend access to its most advanced Claude models, Fable 5 and Mythos 5, for any foreign national, irrespective of their location or employment status with Anthropic. The administration cited “national security authorities” as the basis for this action, although specific concerns were not detailed by Anthropic.

This development comes shortly after Anthropic CEO Dario Amodei published an essay advocating for “serious and binding regulation of AI,” including mechanisms to halt the release of models deemed unsafe. Amodei articulated a vision where, akin to aviation safety standards, frontier AI models should undergo rigorous technical testing and auditing, with their deployment contingent upon meeting high safety benchmarks.

“Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety,” Amodei wrote, underscoring his commitment to a regulated AI ecosystem.

This is not the first instance of Anthropic actively encouraging enhanced government oversight. Since its inception in 2021, following its spin-off from OpenAI, Amodei and his executive team have been vocal proponents of AI regulation, lending their support to legislative efforts at both state and federal levels. The company even lauded a recent executive order on AI signed by President Donald Trump as an “important step” towards responsible AI development.

However, Friday’s directive represents a more forceful form of oversight than Anthropic had envisioned. The company expressed its disagreement with the administration’s assessment, characterizing the disruption as a “misunderstanding” rather than a justified recall. Anthropic stated in a blog post that while they believe the government should have the authority to block unsafe AI deployments, such actions should adhere to transparent, fair, and technically grounded processes, principles they argue were not met by this directive.

Senior Anthropic executives have reportedly traveled to Washington D.C. to engage with administration officials, with the company indicating that both parties are working towards a swift resolution. The timeline for the models’ reinstatement remains uncertain.

Government Mandate vs. Voluntary Cooperation

The recent export control directive stands in stark contrast to the voluntary approach outlined in President Trump’s executive order on AI, issued just ten days prior. That order requested companies to voluntarily submit their AI models for government assessment before full release and tasked officials with developing review frameworks within 60 days.

Sources suggest that Friday’s action was prompted not by the executive order itself, but by discussions between Amazon CEO Andy Jassy and U.S. officials, including Treasury Secretary Scott Bessent. According to reports, Jassy informed the administration that Amazon’s researchers had identified potential security risks with Fable 5, capable of facilitating cyberattacks.

Daniel Remler, a senior fellow at the Center for a New American Security, commented on the shift in government approach. “This sure looks mandatory if there are going to be consequences for not doing what the government says,” Remler noted, highlighting the authoritative nature of the directive compared to the executive order’s voluntary framework.

Fable 5 and Mythos 5 are built upon the advanced capabilities of the Claude Mythos Preview, released in April. This earlier iteration, designed to excel at identifying software vulnerabilities, was initially rolled out to a select group of companies as part of Anthropic’s Project Glasswing, a cybersecurity initiative. While Mythos 5 remained largely restricted, Fable 5 was made available to enterprise clients and paid subscribers, a broader release enabled by newly implemented safeguards aimed at blocking responses in sensitive areas like cybersecurity and biology.

A source familiar with the discussions indicated that Anthropic had engaged with government agencies for pre-release testing and had secured approval for deployment. This source further stated that there was no prior communication from the government regarding national security threats.

“To go from an environment in which we had no controls on model access at all to basically using this tool to take down one company’s model in a few short hours is pretty remarkable,” Remler observed, underscoring the swift and decisive nature of the government’s intervention. He also suggested that in the current volatile policy landscape, such “reactive regulation” might be a foreseeable outcome for a rapidly evolving technology.

Pivotal Moment for Anthropic Amidst IPO Frenzy

This regulatory challenge surfaces at a crucial juncture for Anthropic, as both the company and its main competitor, OpenAI, have recently filed confidential IPO prospectuses. The burgeoning investor appetite for AI ventures, exemplified by SpaceX’s robust post-IPO performance, signals immense market potential. Industry observers are closely watching to see how these upcoming AI IPOs will be received.

Amodei has found allies in prominent industry figures. Alex Stamos, Chief Product Officer at Corridor and former Security Chief at Facebook, penned an open letter co-signed by over 150 executives and technical leaders, urging the lifting of the export control directive. Stamos emphasized the need for AI regulations to be “based upon science” and be transparent, citing a lack of clear guidelines for companies like Anthropic.

“Those rules need to be written down and transparent. That has not happened,” Stamos stated. “There’s nothing that Anthropic or anybody else can look at to say what can and can’t I do.” He expressed concern that businesses might face arbitrary shutdowns if they fall afoul of the administration’s policies.

Anthropic’s relationship with the government has been complex. Earlier this year, the company was embroiled in a significant dispute with the Department of Defense (DOD), which designated Anthropic’s Claude models as a “supply chain risk” in March. This designation requires defense contractors to certify that they will not use Claude models in their military work. Defense Secretary Pete Hegseth alluded to this ongoing tension, suggesting that the current directive validates the DOD’s earlier decision.

Anthropic has initiated legal action against the Trump administration to challenge the DOD’s supply chain risk designation, a case that is currently ongoing.

Stamos believes the administration harbors a particular issue with Anthropic, one of the few prominent AI companies to openly challenge its policies. “If they were fairly enforcing these rules, they would have to enforce them against OpenAI and Google as well,” he remarked, suggesting potential selective enforcement.

Venture capitalist David Sacks, who previously advised the Trump administration on AI and crypto, has been a vocal critic of Anthropic. Sacks accused the company of employing a “sophisticated regulatory capture strategy based on fear-mongering” after an executive published an essay on AI risks. Following the export control directive, Sacks commented on X that Anthropic’s response did not align with its “AI safety company” brand, adding that “The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved. The ball is in Anthropic’s court.”

The situation highlights the intricate dance between innovation and regulation in the rapidly evolving AI landscape. While Anthropic’s commitment to safety has been a cornerstone of its identity, the government’s recent actions underscore the potential for regulatory interventions, even for companies actively promoting oversight. As Anthropic navigates this challenge, its ability to resolve the current directive and its future engagement with regulatory bodies will be closely watched by investors and industry stakeholders alike, particularly as it prepares for its potential public debut.