Summary: Bloomberg broke the news on April 10, 2026: officials in the Trump administration are pushing Wall Street’s biggest banks to put Mythos, Anthropic’s new model, through its paces — the very model the company had gated off only a week before, on the stated grounds that it was too good at finding security vulnerabilities. A TechCrunch follow-up on April 12 seized on the mismatch between this and the executive branch’s own public stance — on March 5 the Pentagon put Anthropic on its supply-chain risk list, and on March 9 Anthropic answered with a lawsuit against the Defense Department.
Bloomberg broke the news on April 10, 2026: officials in the Trump administration are pushing Wall Street’s biggest banks to put Mythos, Anthropic’s new model, through its paces — the very model the company had gated off only a week before, on the stated grounds that it was too good at finding security vulnerabilities. The report adds that, in the week of April 10, Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell brought bank executives together for a meeting built around one message: go test this model.
Every institution named sits at the center of Wall Street. JPMorgan Chase occupies a unique spot — it is Mythos’s sole initial launch partner and the only organization with formal access — while Goldman Sachs, Citigroup, Bank of America, and Morgan Stanley are all reported to be testing the model. For a government to step in and personally promote a commercial model built by one private company is all but unheard-of in the history of US financial regulation — and the company behind this model is, at this very moment, a plaintiff suing the Department of Defense.
Inside the Meeting
As Bloomberg tells it, Bessent and Powell were the conveners, senior bank executives filled the room, and the goal was to give the financial sector direct experience of what Mythos can do in vulnerability detection. The calendar is part of the story: Anthropic announced Mythos on April 7, and within days the administration’s most senior economic officials were recommending the model to the banking world. A TechCrunch follow-up on April 12 seized on the mismatch between this and the executive branch’s own public stance — on March 5 the Pentagon put Anthropic on its supply-chain risk list, and on March 9 Anthropic answered with a lawsuit against the Defense Department. When one arm of Washington puts a vendor on a risk list while the Treasury Secretary and the Fed Chair talk it up to bankers, the government’s position on this one AI company has stopped being a single policy and become two signals running on parallel tracks.
The Reason Behind the Gated Rollout
Access was restricted from day one. Anthropic’s own account: the model is partly “too good at finding security vulnerabilities” — this despite no cybersecurity-specific training. The story landed two ways with observers: one camp saw hype, the other saw “a smart enterprise sales strategy,” their reasoning being that scarcity on its own sets prices and picks the customer base. With government officials now nudging banks toward the model, that safety narrative has gained an official endorsement. The Financial Times, separately, reported that UK financial regulators are debating the risks Mythos brings — meaning the model now sits on the regulatory radar on both sides of the Atlantic.
Suing and Partnering at the Same Time
At the Semafor World Economy Summit on April 14, co-founder Jack Clark said outright that Anthropic had briefed the government on Mythos: “So absolutely, we talked to them about Mythos, and we’ll talk to them about the next models as well.” In his reading, “the government has to know about this stuff,” and he played down the clash with the Defense Department as a “narrow contracting dispute.” Those lines capture the shape of frontier-lab–Washington relations in 2026: inside a single month, you can take the Pentagon to court over a procurement call while walking the rest of the administration through your most sensitive model. Cooperating and fighting are no longer mutually exclusive — the two are handled on separate tracks.
Clark also took the same stage to lower the volume on Anthropic’s earlier job warnings. CEO Dario Amodei had raised the specter of Depression-era unemployment; Clark reports seeing instead, in certain industries, “some potential weakness in early graduate employment,” and his advice to students is to choose majors that “involve synthesis across a whole variety of subjects and analytical thinking about that.” Workforce-policy watchers should note the recalibration: the lab operating nearest the capability frontier now frames the jobs question as entry-level roles thinning out, not the workforce being replaced wholesale.
Takeaways for Finance and for Builders
Three things stand out. First, model procurement has picked up a political charge: once the Treasury Secretary and the Fed Chair vouch for a model in person, the neutral default in bank technology selection is gone, and vendor-governance processes have to read policy winds next to security reviews. Second, restricted release is hardening into a distribution playbook for high-end models — scarcity turns up the volume on the safety story while sifting prospective buyers toward government-grade clients. JPMorgan entering first, with the other four bulge-bracket banks following in to test, is that filter doing its work in real time. Third, cybersecurity is the next arena for governance fights: a vulnerability finder that strong is a defensive tool inside the financial system, but who may hold that access, and who answers when it fails, are questions regulators in Washington and London will have to take on.