Anthropic Mythos sparks alarms

Anthropic’s new Mythos model set off emergency talks in Washington and on Wall Street after officials warned it could sharply increase cyber risk. (bloomberg.com) Bloomberg reported that Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell summoned major bank leaders to an urgent meeting on April 9, 2026. The discussion focused on whether Anthropic’s latest model could make it easier to find and exploit software flaws. (bloomberg.com) Anthropic described Claude Mythos Preview on April 7, 2026 as its “most capable frontier model to date” and said the system showed a large jump over Claude Opus 4.6 on coding and agentic tasks. The company also published a system card and a separate security write-up detailing stronger performance on cyber evaluations. (anthropic.com) In plain terms, the concern is that a model built to understand and rewrite complex code can also act like a very fast bug hunter. Anthropic said Mythos Preview can find higher-severity vulnerabilities, reverse-engineer exploits on closed-source software, and turn some known-but-unpatched flaws into working exploits. (red.anthropic.com) Anthropic did not release Mythos as a normal open product launch. Instead, it said it was limiting access and starting Project Glasswing with partners including Amazon, Apple, Microsoft and Cisco to use the model for defensive security work. (bloomberg.com) (anthropic.com) The banking angle is straightforward: large lenders run sprawling code bases, payment rails and trading systems that already face constant attack. The Federal Reserve has said cybersecurity threats are increasing and that supervision covers both individual firms and risks to the financial system as a whole. (federalreserve.gov) The U.S. meeting did not stay a U.S. story for long. Bloomberg reported that the Bank of Canada met with major lenders on April 10, and that the Bank of England was preparing its own discussions with financial institutions a day later. (bloomberg.com 1) (bloomberg.com 2) At the same time, U.S. officials were not treating Mythos only as a threat. Bloomberg reported that banks were being urged to test the model internally to detect weaknesses, with JPMorgan Chase named as part of the initiative and Goldman Sachs, Citigroup and Bank of America among firms that had access or expected access. (bloomberg.com) Anthropic’s public position is that stronger models can raise both offensive and defensive capabilities at once, which is why it paired the release with restricted access, evaluations and security partnerships. The next question for regulators and banks is whether those guardrails can move as fast as the model itself. (anthropic.com 1) (anthropic.com 2)