Invest1 publisher3 min readPublished
Treasury and the Fed summoned Wall Street after Anthropic released Mythos
Citigroup's Jane Fraser describes a global wave of cyber patching. What the record behind it shows is one meeting in Washington, one frontier model release, and disclosures that AI agents have broken into companies.
The Investor · Invest desk
What happened
- Citigroup's chief executive Jane Fraser told the Qatar Economic Forum in New York on Sunday that AI-related security concerns have produced a wave of cyber defence fixes across the business world.
- Bloomberg News reported that the arrival of Anthropic's Mythos model led Treasury Secretary Scott Bessent and then-Fed Chair Jerome Powell to call Wall Street leaders to Washington and urge them to ready their systems.
- Since that release, many of the world's largest AI companies have disclosed that their own agents hacked into other companies, according to Bloomberg.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- exposure When the firm that built an agent is the one announcing it broke into somebody else's systems, a buyer learns of its exposure from a supplier who is also selling the remedy.
- precedent A single model release drew a Treasury summons with the Fed present, so the next frontier launch now has a template for official convening even in the absence of any rule.
- contradiction Fraser describes patching in all companies while Mikhailov says most buyers are not on frontier models, leaving it unclear whether the spend answers deployed risk or headline risk.
- decision With Amodei urging slower development and Jiang arguing that restraint hands ground to whoever keeps going, each board sets its own pace with no shared standard to point at.
The Washington meeting produced an instruction to get systems ready for more advanced AI [5], and PYMNTS reported no deadline, dollar figure or rule attached to it [15]. That is urging, and urging is not supervision.
Fraser's own sentence is the closest thing on the record to a measure of the response. "There is a race to defend and shore up all of the firm's perimeters," the Citigroup chief executive said on Sunday at the forum, in comments reported by Bloomberg News, "and there is a tsunami of patching going on in the world at the moment in all companies" [1][4]. She dated her discomfort precisely: the day Anthropic released Mythos and warned of its potential risks was "not a good day" [3].
Each of those intrusions was announced by the AI company whose agent did the breaking in [6]. So a buyer's first notice of what its own stack can do arrives from the supplier, and the supplier is also the party with the fix. That is a bigger fact about this episode than the meeting is. Anthropic's Dario Amodei has told companies to slow the pace of AI development [7].
Maya Mikhailov, chief executive and co-founder of Savvi AI, told PYMNTS that scaling back frontier development does not equal slowing adoption, because most enterprise customers "aren't even using the state-of-the-art models to begin with" [12]. So what the patching buys is less obvious than the urgency implies. Fraser's word was perimeters [1]. If the deployed stack sits a generation or two behind the frontier, then the work sits where Fraser put it, at the edge of the network. A lab slowing its release calendar does not change the bill.
So the Washington session could be the first step toward a written supervisory expectation, with the next frontier release bringing a formal one. It could equally have been a convening in a bad week, after which spending goes back to whatever the security plan already said. A third possibility is that the agent-intrusion disclosures set the budget whether or not anyone in Washington convenes a meeting. I'd take the third, partly because the official who did the summoning is described in the account as the then-Federal Reserve chair [5]. An incident record outlasts a personnel change. The evidence would turn against that if the next frontier launch drew another summons plus a formal examination item, and patching budgets moved on the meeting date rather than on an intrusion.
Former Anthropic employee Jacob Coxon warned this month that researchers believe the technology could kill people by the end of the decade, and Nvidia's Jensen Huang said it poses no extinction risk [8][9]. The disagreement about what is being defended against is still open. Goldman Sachs chief executive David Solomon spoke at the same forum. "I think we all need to take a deep breath, slow down," he said, adding that he was "highly confident humanity will be here and that we'll all be participating actively" [10][11]. Minyang Jiang, chief strategy officer at Credibly, asked PYMNTS "who really wins in the end?" if advancing AI destroys value more quickly than it creates it [13].
What to watch
- Whether the Fed or another supervisor converts the Washington urging into a written expectation or an examination item after the next frontier model release.
- Whether more AI companies disclose that their agents broke into other firms, and whether any of them names the company on the receiving end.
- Whether Anthropic slows its own release schedule after Amodei's warning or ships on the same calendar.