Anthropic Sends Staff to D.C. Amid AI Model Export Ban

AI startup Anthropic has dispatched key technical personnel to Washington, D.C., to engage in urgent negotiations with officials from the Trump administration. This move follows the implementation of export restrictions that have forced the company to halt access to its most advanced AI models.

Reports suggest that Anthropic is locked in intensive discussions with the Trump administration to address the new export restrictions affecting its flagship AI systems, Fable 5 and Mythos 5. In a bid to find a resolution, the company sent its leading tech experts for a weekend meeting aimed at regaining access to these sophisticated models.

Recently, the Trump administration instituted a ban limiting foreign access to Fable 5 and Mythos 5, leading Anthropic to completely cut off access to comply with these regulations. The ban has impacted various foreign governments, companies, and some of Anthropic’s own employees who are not U.S. citizens.

On the following day, high-level discussions took place regarding a simplified version of Mythos, called Fable 5, designed for public use. Key figures in these talks included Secretary of Commerce Howard Lutnick, National Cyber Director Sean Cairncross, and several leading members of the Anthropic team.

The export restrictions have drawn immediate criticism from the cybersecurity community. Prominent experts issued a letter urging the administration to reconsider its decision, arguing that the ban undermined U.S. AI leadership and heightened market uncertainty without adequately justifying the associated risks.

Despite the ongoing tensions, sources indicate that both the Trump administration and Anthropic have a shared interest in resolving the issues and restoring access to the advanced AI models, though details on any potential compromise are unclear. Officials deem collaboration between Anthropic’s technical staff and government security experts essential in finding a way forward.

This situation marks the latest in a series of disputes between the Trump administration and leading AI labs regarding the regulation and use of cutting-edge technology. Recently, governments have increased their efforts to monitor the rapidly changing AI landscape.

Anthropic has consistently asserted that it prioritizes safety and has robust safety measures in place. The startup is preparing for a potential public offering later this year and has faced ongoing disagreements with the Pentagon over safeguards particularly related to military applications and various policy matters.

Currently, Anthropic is collaborating with the White House on deploying its Mythos model, which is being used by about 200 select entities, including businesses and researchers, to identify and address software vulnerabilities in sectors like healthcare and finance. This week, the company introduced Fable, a more restricted version of Mythos, which has been equipped with extensive safety features.

Interestingly, the government restrictions do not extend to Mythos Preview, which remains accessible to a specific group of partners.

The controversy surrounding Fable began last week when researchers from Amazon demonstrated the ability to bypass certain safeguards, raising alarms among White House officials. They showed that altering the query methods could reveal information on security vulnerabilities in multiple software applications.

While cybersecurity experts agree that the findings may help those tasked with network protection, they clarify that the study does not suggest that Fable is generating riskier outputs, like exploit software for cyberattacks.

Amazon CEO Andy Jassy shared these findings with U.S. officials, leading to a meeting among White House personnel to decide on the next steps, resulting in a request for Anthropic to restrict access to the model. President Trump approved the restrictions and assigned Lutnick to spearhead the response.

Discussions between Anthropic’s leadership and key government officials about the company’s safety measures took place, raising the potential for further export restrictions. A source noted that the company is feeling the pressure of quickly complying with the new rules.

That evening, Anthropic complied with the administration’s directive and ended all access to Fable 5 and Mythos 5.

In response to the concerns raised by Amazon, Anthropic stated that the identified vulnerabilities are basic and can be found by other publicly available models, arguing that these do not represent a complete security breach—a viewpoint shared by some experts familiar with the research.

To navigate this conflict, Anthropic has sent a team of senior engineers to Washington, including top security researcher Nicholas Carlini and lead risk model evaluator Logan Graham, to engage in discussions with government security officials. Administration sources have indicated that such meetings are vital to resolution efforts.

Resolving this situation is crucial for Anthropic as it aims for a $1 trillion IPO, especially following the recent milestone IPO by SpaceX.