Fed sought access to Anthropic’s Mythos AI after warning banks

The Federal Reserve warned top U.S. banks in April that Anthropic’s Claude Mythos Preview could create a serious cybersecurity risk, but the central bank did not have access to the model for at least three months afterward, CNBC reported. For investors following banks and AI companies, the episode shows how quickly advanced AI tools are becoming part of financial-sector risk management.

Claude Mythos Preview is an artificial intelligence model from Anthropic that the company has said is strong at finding software flaws and security weaknesses. That ability can help institutions fix vulnerabilities, but it also raises concern if similar capabilities are used to identify targets before defenses are updated.

According to CNBC, the Fed and Treasury Department brought together chief executives from major banks in April to discuss the model and its potential threat to large financial institutions. The meeting took place while Jerome Powell was leading the Fed and Treasury Secretary Scott Bessent was involved, CNBC reported.

The model was distributed through Anthropic’s Project Glasswing, a cybersecurity effort that initially gave access to about 50 organizations. Anthropic named only some participants at the time, including JPMorgan Chase, Amazon, Apple and Google, CNBC reported.

Warsh said the Fed was still seeking access

Fed Chair Kevin Warsh told the Senate on July 15 that he was still pushing for access to Mythos and other advanced AI systems, according to CNBC. The central bank declined to comment to CNBC, and Anthropic did not immediately respond to CNBC’s request for comment.

“We are not the deciders as to who has access, but I have not been shy in sharing my views with authorities across the government about the vulnerabilities, and have been asking for access not just for the Federal Reserve but for other institutions to a whole range of these new artificial intelligence models so that they can protect themselves,” Warsh told Sen. Jack Reed, D-R.I., according to CNBC.

Warsh also said he did not want the issue limited to one product. “As these new models find their way more broadly, our banking system and frankly, the Federal Reserve needs to do all we can to patch any vulnerabilities that we have,” he said, according to CNBC.

CNBC reported that the Fed had not secured access as of July 15. Whether the central bank obtained access after that date was not known.

Access widened, then got tangled in policy

Anthropic expanded Project Glasswing in June to include more than 150 organizations across 15 countries, CNBC reported. Daniel Newman, CEO of research firm the Futurum Group, told CNBC he was surprised the Fed had not been included, given the central bank’s role in the financial system.

The rollout also became tied to federal export controls. In June, Anthropic said it disabled access to Mythos 5 and Fable 5 to comply with a government directive citing national security authorities, CNBC reported. Commerce Secretary Howard Lutnick later allowed Mythos access to resume for certain “trusted partners,” according to a letter reviewed by CNBC, and the export controls were later lifted entirely.

The policy backdrop has shifted under President Donald Trump, who signed an AI executive order in June, CNBC reported. The Center for AI Standards and Innovation has also seen leadership turnover, with CNBC confirming that Chris Fall resigned as director three months after being selected for the role.

Competition is adding pressure. CNBC reported that Chinese open-weight models, meaning models whose internal parameters are made more available for outside use, have gained attention against U.S. systems from companies including OpenAI and Anthropic. Moonshot AI’s Kimi K3 model outperformed those companies on some industry benchmarks, CNBC reported.

This story draws on original reporting from CNBC.