On April 8, U.S. Treasury Secretary Bessent and Federal Reserve Chair Powell urgently convened the CEOs of six systemically important banks at the Treasury headquarters in Washington to discuss the potential cybersecurity risks posed by Anthropic’s new model, Mythos. According to Bloomberg, representatives from Citigroup’s Fraser, Morgan Stanley’s Peck, Bank of America’s Moynihan, Wells Fargo’s Schafer, and Goldman Sachs’ Solomon all attended. JPMorgan Chase’s Dimon “was unable to attend due to unforeseen circumstances.”
This is not a routine meeting of financial regulators. The last time the Treasury Secretary and the Chair of the Federal Reserve jointly summoned bank CEOs to the same room was on October 13, 2008. Paulson and Bernanke introduced the $250 billion TARP capital injection program to halt an ongoing collapse of the financial system. This time, the trigger is not the market—it’s an AI model.
This meeting was the third wave of federal regulatory actions against Anthropic within 44 days.
The people in the lab and the people in the meeting room hardly overlap.
Mythos is not freely scattered across the world. On April 7, Anthropic officially released the Mythos Preview, alongside the announcement of Project Glasswing, a defensive cybersecurity initiative that restricts model access to 12 partner organizations. These 12 include Amazon, Apple, Microsoft, Google, NVIDIA, CrowdStrike, Palo Alto Networks, and a financial institution, JPMorgan Chase.
The problem is that the people called in to discuss risk are almost never the same as those who receive the defensive tools.
The five bank CEOs present represented institutions that collectively manage approximately $9 trillion in assets. None of them are on the Glasswing list, and none have access to Mythos. According to Bloomberg, they were called in to “ensure banks understand the potential future risks posed by Mythos and similar models and are taking defensive measures.” Yet the core tool for those defensive measures—Mythos itself—is inaccessible to them.
JPMorgan Chase, the only institution appearing on both lists, had its CEO, Dimon, absent.
According to Anthropic’s official website, JPMorgan Chase is the only financial institution among Glasswing’s 12 founding partners. The bank spends approximately $600 million annually on cybersecurity and employs around 3,000 cybersecurity personnel, according to SecurityWeek. On the same day—April 8—that Bassett and Powell held an emergency meeting, JPMorgan Chase analysts released a research report expressing bullish views on CrowdStrike and Palo Alto Networks, citing the formation of Glasswing as the key rationale.
On one side are five banks summoned to discuss threats, armed with no tools. On the other side is the single bank that received the tools—its CEO didn’t attend the meeting, and analysts are still issuing buy ratings to partners. Glasswing has created not just a technological barrier, but an information barrier.
What happened in 44 days?
When you line up the federal regulatory actions Anthropic has faced over the past six weeks, you see an unprecedented chain reaction in both speed and direction.
On February 24, Secretary of Defense Hegseth issued an ultimatum to Anthropic CEO Amodei: accept the "any lawful use" clause by 5:01 p.m. on February 27, or face consequences. The core controversy centers on two usage restrictions Anthropic retained in its contract with the Pentagon, prohibiting the use of Claude for mass surveillance and fully autonomous weapons systems. According to CBS News, Hegseth stated, "America's warriors should never be held hostage by the ideology of Big Tech." Meanwhile, Pentagon Chief Technology Officer Emil Michael said, "At some point, you have to trust your military to do the right thing."
Anthropic refused. On February 27, Trump ordered federal agencies to discontinue the use of Anthropic products. On the same day, Hegseth listed Anthropic as a "supply chain risk."
For the next 44 days, three federal agencies simultaneously took actions against Anthropic that were mutually contradictory.
On March 26, Judge Lin of the U.S. District Court in San Francisco issued a 43-page preliminary injunction, characterizing the Pentagon’s actions as “disturbing” and potentially constituting retaliatory punishment against Anthropic, temporarily blocking the enforcement of the supply chain risk designation. According to CNN, the judge wrote in the ruling that the government’s actions “likely violate the law.”
On April 8, the DC Court of Appeals denied a stay of the Pentagon’s determination, effectively overturning the protection granted by the San Francisco court. On the same day, Bessent and Powell held that emergency meeting.
One court is protecting this company, another is supporting sanctions against it, while the Treasury and the central bank are discussing it from the perspective of financial stability as a systemic risk. The same entity, within the same federal system, is simultaneously regarded as an innovator requiring protection and a threat requiring mitigation.
Who previously used the "Supply Chain Risk" label?
"Supply chain risk" is not an ordinary administrative designation. It was originally designed by the Federal Communications Commission (FCC) to address security threats posed by foreign telecommunications equipment.
In 2019, Congress and the FCC first initiated this designation process against Huawei and ZTE due to their ties to the Chinese government, legal obligations under Chinese law requiring them to engage in espionage, and known security vulnerabilities in their equipment. In June 2020, the FCC issued a formal designation order. In 2021, Hikvision, Dahua, and Hytera were added to the list. In 2022, Russia’s Kaspersky was also included.
According to FCC public records, prior to Anthropic, all entities on this list were foreign companies, all of which were identified as posing technical security risks.
On February 27, 2026, Anthropic became the first U.S.-based company to be listed under “supply chain risk.” According to TechPolicy.Press, the designation was not due to a technical security vulnerability, but rather due to a breakdown in contract negotiations, as Anthropic refused to remove restrictions on AI usage. Security policy researchers at Northeastern University assessed that this precedent could have a chilling effect on the entire AI industry, as it implies that refusing to comply with military usage conditions may trigger the highest level of supply chain sanctions.
A sanction tool designed for Huawei has, for the first time, been used to sanction a U.S. company that refused to allow its AI to be used for unlimited military purposes. The tool remains the same, but the logic behind its use has completely changed. On May 19, the D.C. Court of Appeals will hear oral arguments in this case.
When an AI model is simultaneously the best defensive tool and the greatest systemic threat, existing regulatory frameworks were never designed to address this dual nature of "capability as risk," and the process of filling this regulatory gap is unfolding in real time—at a rate of one federal-level incident per week.