On Tuesday, April 7, 2026, Federal Reserve Chair Jerome Powell and U.S. Treasury Secretary Scott Bessent convened an unscheduled meeting at the Treasury Department with the chief executives of Bank of America, Citigroup, Goldman Sachs, Morgan Stanley, and Wells Fargo. The bank heads were already in Washington for a Financial Services Forum board meeting; they had been together at a dinner the night before. Then they were called in. JPMorgan’s Jamie Dimon was the only major bank CEO who could not attend.
The subject was a single AI model.
Earlier that same week, Anthropic had released Claude Mythos Preview in a deliberately limited capacity, citing concerns that hackers could exploit its capabilities. The model has, by Anthropic’s own disclosure, both offensive and defensive cyber applications. The release was paired with a cybersecurity initiative called Project Glasswing, with JPMorgan, Apple, Google, Microsoft, and Nvidia as initial partners. Anthropic had briefed senior U.S. government officials before launch, including the Cybersecurity and Infrastructure Security Agency and the Center for AI Standards and Innovation.
That sequence — limited release, regulator briefing, and an unscheduled meeting between the two most powerful monetary officials in the United States and the heads of the country’s largest banks — is the news. It was first reported by Bloomberg and the Financial Times, then confirmed by CNBC on April 10. The Fed and Treasury declined to comment publicly.
We want to be careful here. The headline is dramatic, and there is a temptation to either dismiss it as theater or inflate it into prophecy. We are doing neither. We are reading it for what it is: a signal about how the regulators of the U.S. financial system now think about a category of risk that did not exist on their map two years ago.
What the Meeting Actually Tells Us
Powell and Bessent do not pull bank CEOs into unscheduled sessions for routine concerns. The structure of the meeting matters as much as the topic. The bank chiefs were already in town. The convening was added to an existing window. It was confidential enough that the participants asked not to be named, and significant enough that two of the most respected financial publications in the world both broke it within the same news cycle.
This is not the first time AI cyber capability has surfaced as a financial-system concern, and that is the point. In November 2025, Anthropic itself disclosed that a Chinese state-linked group had used its Claude models to automate hacks against government and corporate targets. In late March 2026, Fortune surfaced a draft Anthropic blog post that revealed the advanced cyber capabilities and risk profile of the model that would become Mythos; cybersecurity stocks slumped on the news before the official release. By the time the Tuesday meeting was called, the regulators were not reacting to a surprise. They were reacting to a pattern they had been tracking for months — and they decided that pattern now warranted a direct, off-calendar conversation with the people who run the country’s largest banks.
“That is the regime change. AI cyber capability has formally entered the same systemic-risk conversation that, until recently, was reserved for credit-cycle stress, counterparty failure, and sovereign liquidity events.”
The category has expanded. The list of things that can break the financial substrate has grown by one.
The CI Mavericks Reading
We have written before — repeatedly — that the architecture our members depend on is more brittle than its participants admit. Correspondent banking is contracting. Account access for non-resident structures has narrowed every year for a decade. The dollar-denominated digital infrastructure that moves and stores most of the world’s wealth runs on a small number of interconnected systems, and the assumption that those systems are safe rests on the assumption that the people trying to break them are not catching up. The Tuesday meeting is the regulators acknowledging, in private, that the people trying to break them may now be catching up faster than the defenders can react.
Three things follow from that.
First, concentration risk is no longer just credit and counterparty risk. When you hold wealth at a major institution, you have always borne the credit risk of that institution and the counterparty risk of its trading book. You now also bear a category of cyber-substrate risk that even the regulators do not yet know how to price. We are not predicting a breach. We are observing that the people whose job it is to prevent one have just signaled that they cannot rule one out.
Second, alternative means of moving and storing worth become more important precisely because the conventional rails are now formally on the regulator watchlist. This is not a panic argument. It is a diversification argument, and it is the same one we have made about jurisdictional concentration, currency concentration, and custody concentration for years. Physical custody of hard assets, non-bank holding structures, multi-jurisdictional architecture, and operating businesses with real economic substance are not exotic positions. They are insurance against the kind of substrate failure that the Tuesday meeting was convened to discuss.
Third, this is why the CI Mavericks portfolio is built the way it is. Our segregated portfolios hold real estate, agriculture, energy, and treasury — assets that exist whether or not the digital rails are working. We hold physical gold in multiple jurisdictions. Our operating exposure runs through Argentine and Cayman businesses with on-the-ground substance. The JV → SPC architecture itself is a sovereignty-diversification tool: it spreads the legal, custodial, and regulatory surface area across multiple jurisdictions so that no single substrate failure removes our access to our own capital.
We did not build that structure in response to the Mythos meeting. We built it because we have been reading the same signals the regulators are now reading — slowly, in public, and with a delay measured in years.
The Conference Conversation
This is the conversation Day One of our Inaugural Strategic Conference is built around. Session 2, Moving Wealth Across Boundaries, addresses the contracting universe of correspondent banking and the practical alternatives to moving wealth through the conventional financial system. Session 3, Hard Assets Outside the Financial System, addresses precious metals and physical custody as a portfolio response to monetary and substrate risk. Both sessions were on the agenda before the Tuesday meeting in Washington. The agenda did not need to be updated.
The Bottom Line
We do not predict the timing of system failures. We do not know whether Mythos, or the next model, or the one after that, will be the capability that finally breaks something the regulators cannot fix in time — or whether Mythos will be used as the excuse to bring down the system so that they may usher in the monetary reset. What we know is that the people whose job it is to fix those things have now told their largest counterparties, in private, that the risk is real enough to convene about.
“We plan. We invest for the long term. And we hold positions that do not depend on any single system continuing to work — because the regulators have just confirmed, in the most credible way they know how, that no system is permanent.”
Published for informational and educational purposes only. Does not constitute legal, tax, or investment advice. CI Mavericks Advisory Services maintains active positions in the asset classes and jurisdictional structures discussed. CI Mavericks uses Anthropic’s Claude models as a tool in research and content drafting and discloses this as standard editorial practice. Readers should consult qualified financial, legal, and tax advisors before making any decisions based on this content.
Sources: CNBC (Samantha Subin and Hugh Son, April 10, 2026, updated April 13, 2026); Bloomberg; Financial Times. All facts regarding the meeting, attendees, and Mythos release are drawn from these public reports.