Anthropic CEO Dario Amodei is set to meet White House chief of staff Susie Wiles on Friday. The meeting centers on Anthropic’s dispute with the Pentagon over limits on military use of its AI systems. It also comes as federal agencies, banks, and regulators test the company’s new Mythos model.
White House meeting marks new step in Anthropic’s Pentagon dispute
Amodei’s visit to the West Wing follows months of tension between Anthropic and the Pentagon. The company is suing after the Pentagon blacklisted Anthropic over usage restrictions. The dispute began after Amodei refused to allow unrestricted use of Anthropic’s AI models. That decision triggered a legal and political fight inside Washington.
According to Axios, the Trump administration sees strategic value in Anthropic’s latest system. Officials view Mythos as a powerful model with advanced cyber capabilities. A source close to negotiations said the government should not deny itself the model’s technological advances. The source added that such a move “would be a gift to China.”
The Friday meeting is designed to reopen channels after talks with the Pentagon cooled. It also signals renewed contact between Anthropic and senior officials. This is the second high-level meeting between Amodei and a top Trump official this year. The earlier meeting took place in late February. At that time, Defense Secretary Pete Hegseth gave Amodei a deadline to accept Pentagon terms.
However, Anthropic rejected those terms, and the conflict continued. Since then, both sides have remained locked in a legal and political standoff. Some administration officials now see that fight as counterproductive. Parts of the U.S. intelligence community are already testing Mythos in controlled settings. The Cybersecurity and Infrastructure Security Agency is also evaluating the model. Treasury officials and other agencies have also shown interest in access. That agency demand has increased pressure for a policy resolution.
What Mythos is and why agencies are testing it
Anthropic released Mythos as a cyber-focused model within its Claude family. The company says it can identify and exploit weaknesses in digital systems. The company also released a new version of Claude Opus on Thursday. Anthropic said that the release would let researchers test Mythos-like capabilities on weaker systems.
Anthropic has claimed that Mythos exposed security flaws in critical operating systems. It also said the model found weaknesses in financial systems and web browsers. Those claims have drawn attention because the model has not undergone wide industry testing. Some cyber experts have questioned how far the concerns should extend.
The UK’s AI Security Institute received early access to a preview version of Mythos. It was later published as the only independent report cited in the provided material. Its researchers said Mythos could exploit systems with weak security posture in undefended environments. However, they did not describe Mythos as a major jump over Opus 4.
The report stated, “Our testing shows that Mythos Preview can exploit systems with weak security posture.” It added that more models with similar capabilities will likely follow. That finding placed Mythos within a wider trend in cyber-focused AI development. It also gave public institutions a reference point before any wider rollout.
The report also showed that outside testing remains limited at this stage. As a result, much of the official response still relies on preview access. The text also notes that AI developers have made similar release decisions before. In 2019, OpenAI staggered GPT-2’s release while citing misuse concerns. Critics have argued that such warnings can also build attention around a product launch. Even so, Mythos is already in early testing with public agencies.
Bankers and officials raise concerns as Mythos testing expands
The White House visit comes at a time when bankers are testing new cyber defenses against Anthropic’s latest AI system. According to a report by the BBC, major banks and government agencies received early access to Mythos before any public release. The BBC reported that top bankers would use that access to test their systems. Officials want institutions to find and fix weaknesses before criminal misuse expands.
Barclays chief executive CS Venkatakrishnan told the BBC, “It’s serious enough that people have to worry.” He added that banks need to understand exposed vulnerabilities and fix them quickly. Venkatakrishnan also linked the issue to a more connected financial system. He said the new environment brings both opportunities and vulnerabilities.
Bank of England governor Andrew Bailey also addressed the model’s cyber risk. He said officials must study what the latest AI development could mean for cybercrime. Bailey warned that improved AI tools could make it easier to detect existing vulnerabilities. He added that bad actors could then seek to exploit those flaws. The U.S. Treasury confirmed it had raised the issue with major banks. Officials encouraged those banks to test their defenses before any wider release. Financial industry sources told the BBC that another U.S. AI company may soon release a similar system. That possibility has added urgency to current testing efforts.
James Wise, a partner at Balderton Capital, also commented on the model’s place in the market. He said Mythos is “the first of what will be many more powerful models.” Wise chairs the Sovereign AI unit, which backs British AI companies with government funding. His remarks came as banks, agencies, and regulators continued early Mythos testing. At the same time, officials in Washington are weighing Anthropic’s dispute with the Pentagon. Friday’s White House meeting places that conflict at the center of the latest Mythos response.