Anthropic to meet with Trump administration over Mythos dispute

In a significant development, senior executives from Anthropic are set to engage with officials from the Trump administration in Washington, D.C., on Monday. This meeting aims to address a major dispute involving the artificial intelligence firm and the U.S. government. The conflict escalated after Anthropic received an export control directive on Friday, invoking "national security authorities" and mandating the suspension of access to its latest AI models, Fable 5 and Mythos 5, for all foreign nationals, both domestically and internationally. To comply with this directive, Anthropic took the step of disabling access to these models across its customer base. This situation marks yet another chapter in the complicated relationship between Anthropic and government entities, particularly following earlier tensions with the Department of Defense (DOD). Earlier this year, the DOD classified Anthropic as a supply chain risk, effectively prohibiting defense contractors from utilizing its technology due to perceived national security threats. Defense Secretary Pete Hegseth commented on the recent directive via social media, asserting that the decision to blacklist Anthropic was justified. Amid these developments, Anthropic has also initiated legal action against the Trump administration, seeking to overturn the supply chain designation. Just days prior to the export control order, Anthropic had launched Fable 5 and Mythos 5, after collaborating with government agencies to test these models. A source familiar with the discussions indicated that the company had received prior approval for their deployment. However, the situation took a turn when the government contacted Anthropic on Friday afternoon, demanding the halt of these models due to an unspecified national security concern. A formal letter followed, reinforcing the requirement to suspend access. Notably, prior to this directive, Anthropic had not been made aware of any potential threats. Fable 5 and Mythos 5 are built on Anthropic's Claude Mythos Preview, a sophisticated model recognized for its proficiency in identifying security vulnerabilities in software. Initially, the rollout was limited to a select group of companies as part of a cybersecurity initiative named Project Glasswing, which seemed to foster a positive rapport with the Trump administration. Anthropic has described Fable 5 and Mythos 5 as leading-edge models that excel in various industry benchmarks. While Mythos 5 remains restricted to a chosen group of users, Fable 5 has been made available to enterprise clients and paid subscribers. The broader release was facilitated by the introduction of new safeguards designed to prevent misuse in high-risk domains, including cybersecurity and biology. In its statement regarding the directive, Anthropic expressed its belief that the government's concerns are centered around a "narrow, non-universal jailbreak" risk, where a user might circumvent cybersecurity measures to manipulate the models. "We contend that the identification of a narrow potential jailbreak should not necessitate the recall of a commercial model utilized by millions," Anthropic stated. "If similar standards were applied industry-wide, it could effectively stymie the deployment of new models by all leading providers." Anthropic maintains that this conflict stems from a misunderstanding and is committed to resolving the issue to restore access to its models as swiftly as possible.