Anthropic model sparks alarm
U.S. financial regulators urgently called in Wall Street after an Anthropic model raised serious cyber-risk concerns that officials say could affect markets and infrastructure. The meeting was prompted by reports and social posts flagging the model's risks, signaling regulators are treating some advanced models as systemic cyber issues rather than mere product problems. (x.com)
Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell pulled top Wall Street chiefs into an urgent meeting this week because they think one new Anthropic model could make cyberattacks faster, cheaper, and harder to stop. Bloomberg reported the warning was serious enough that officials treated it as a threat to markets and critical infrastructure, not just a bad product launch. (bloomberg.com) The model at the center of it is Claude Mythos Preview, which Anthropic announced on April 7, 2026. Anthropic said it is not releasing the model to the public and is limiting access to selected partners because of its cyber capabilities. (anthropic.com) Anthropic built Mythos to read software the way a master locksmith reads a lock. In its Project Glasswing post, the company said the model can find and exploit software vulnerabilities at a level beyond almost all human experts. (anthropic.com) A software vulnerability is a hidden mistake in code that can let an attacker break in, steal data, or shut systems down. Anthropic said Mythos has already been used to scan critical software and help defenders patch weaknesses before attackers can use them. (anthropic.com) That is why Anthropic paired the model with Project Glasswing instead of a normal public launch. The company said more than 40 organizations that build or maintain critical software infrastructure will get access for defensive work, and it committed up to $100 million in usage credits plus $4 million in direct donations to open-source security groups. (anthropic.com) The banking meeting happened because banks sit on two kinds of targets at once: money rails and identity data. If a model makes it easier to discover flaws in payment systems, trading software, cloud tools, or employee devices, a single bug can jump from a technical problem to a market problem very fast. (bloomberglaw.com) This did not come out of nowhere. On March 27, 2026, CNBC reported that cybersecurity stocks fell after reports about Mythos spread, showing investors were already gaming out what happens if artificial intelligence starts doing high-end offensive security work at scale. (cnbc.com) Anthropic’s own rollout added to the alarm because the company framed Mythos as a model powerful enough to justify withholding from general release. CNBC reported on April 7 that Microsoft, Amazon, Apple, CrowdStrike, and Palo Alto Networks were among the companies given access through Glasswing for defensive testing. (cnbc.com) What changed this week is who is reacting. When the Treasury Department and the Federal Reserve call in bank chief executives over one model, they are treating advanced artificial intelligence less like a software update and more like a system-wide cyber risk, the way regulators think about stress, contagion, and weak links. (bloomberg.com)
