Hegseth, White House allies intensify attacks on Anthropic

Hegseth, White House allies intensify attacks on Anthropic

Hegseth White House allies intensify attacks - Defense Secretary Pete Hegseth has escalated his criticism of Anthropic, the artificial intelligence company, following the decision to restrict access to two of its models in compliance with a Trump administration directive. This move, made by Anthropic on Friday, came after the government ordered the suspension of all foreign access to the Fable 5 and Mythos 5 models, prompting the company to disable them for all users. Hegseth, who has been a vocal advocate for stricter AI regulations, took to the social media platform X on Saturday to highlight the decision, suggesting it validated his earlier stance on the matter.

Previous Tensions Between Anthropic and the Pentagon

Hegseth’s actions this week are part of a broader pattern of scrutiny against Anthropic, which began earlier this year. At that time, he labeled the company a potential risk to the U.S. supply chain, leading to a ban on the use of its Claude AI chat bot within the Pentagon. This designation was a direct response to concerns raised by Anthropic’s CEO, Dario Amodei, about the model’s potential applications in autonomous lethal weapons and domestic surveillance. The company has since filed a lawsuit against the administration, arguing that the restrictions were overly harsh and not well-justified.

"Three months ago, @DeptofWar expelled @AnthropicAI from its premises permanently. Every passing day shows that this was a wise decision," Hegseth tweeted, reflecting his continued support for the policy.

The administration’s directive was aimed at ensuring that AI models like Fable 5 and Mythos 5 did not fall into the hands of foreign entities, which could be used to compromise national security. Anthropic, however, claims the order was triggered by the discovery of a method to "bypass" the model’s security protocols, a process known as "jailbreaking." According to the company, this technique was demonstrated to reveal a small number of previously identified, minor vulnerabilities. These flaws, while simple, were deemed significant enough to warrant immediate action.

Anthropic’s Response and the 90-Minute Deadline

In a statement, Anthropic emphasized that it had already collaborated with federal agencies to conduct pre-release testing of Fable 5 and received explicit approval before its launch. The company expressed frustration over the abrupt nature of the latest directive, which required them to disable access to the models within a 90-minute timeframe without prior warning. A source close to the company told The Hill that the administration informed them of the requirement just hours before the deadline, leaving little room for negotiation.

"We have completed extensive testing with the relevant agencies and were given the green light to deploy the model. The 90-minute window was unanticipated, and we believe the government’s urgency was driven by a specific concern they uncovered," the source explained.

Despite the quick turnaround, Anthropic maintains that the action was unnecessary and would have a broad impact on the AI industry. They argue that the government could have followed a more transparent process, allowing for a statutory review before implementing such a sweeping restriction. The company also noted that similar vulnerabilities have been found in other publicly available models without the need for a bypass method, suggesting the issue was not unique to their system.

Trump’s Executive Order and Anthropic’s Stance

The controversy over Anthropic’s models is linked to an executive order signed by President Trump in January, which allowed AI labs to voluntarily share their models with the government for up to 30 days before public release. Amodei, who has been a key figure in shaping Anthropic’s public image as a safety-focused research firm, initially supported this policy. However, the latest directive appears to contradict that approach, with Anthropic stating that the government’s action would "essentially halt all new model deployments for all frontier model providers."

"We believe the government should have the ability to block unsafe deployments through a transparent, fair, and technically grounded process. This latest action does not align with those principles," the company stated.

The White House has defended the directive, asserting that it was necessary to address emerging risks in AI technology. David Sacks, a special adviser to Trump on AI and cryptocurrency, criticized Anthropic’s response, calling it inconsistent with the firm’s branding as a leader in safe AI. "The administration hopes Anthropic will fix the safety issue, lift the export control, and allow Fable 5 back into general release," Sacks wrote on X. "It’s baffling that they haven’t prioritized compliance with these safety requests as they claimed."

Support from Congressional Allies

Administration officials are not the only ones backing the directive. Democratic Senator Mark Kelly of Arizona praised the decision on Sunday, urging AI companies to exercise caution when releasing new models. "Anthropic seems willing to work with the federal government to prevent mistakes, such as deploying a model we later regret," Kelly told CBS News host Margaret Brennan during a segment titled "Face the Nation." His comments underscore the bipartisan concern over the potential risks of unregulated AI advancements.

While the directive has drawn criticism from some quarters, it reflects a growing consensus among policymakers that AI companies must be held accountable for their technology’s security and ethical implications. The dispute over Fable 5 and Mythos 5 highlights the tension between innovation and control, as Anthropic strives to balance its ambitions with the demands of regulatory oversight. With the company aiming for a potential $1 trillion valuation by going public later this year, the outcome of this conflict could have significant repercussions for its future and the broader AI sector.