Fed warned banks on cybersecurity but went months without the Anthropic AI model itself
The US Federal Reserve pressed banks to patch cybersecurity vulnerabilities flagged under a project referencing Anthropic's AI, yet the central bank itself lacked access to the Claude Mythos Preview model as of mid-July. Other institutions were already racing to close the gaps it identified.
The Federal Reserve raised alarms about cybersecurity vulnerabilities across the banking system under an initiative referred to internally as Project Glasswing, pointing to risks tied to Anthropic's AI. Yet the central bank itself did not have access to the Claude Mythos Preview model that underpinned the warning as of mid-July, according to CNBC.
Other institutions were already working to patch the vulnerabilities the Fed had flagged, even as the regulator lagged behind on the very tool needed to verify them. Banking industry figures said the gap raises questions about how effectively the Fed can supervise fast-moving AI risks if it cannot access the systems it is warning about.
The episode has reopened debate over how central banks structure data-sharing and access agreements with AI developers. Anthropic has not issued a public comment on the access delay.
Read next
Hungarian prosecutors raid data servers of Orban's Fidesz party, party says
Hungary's former ruling Fidesz party said prosecutors showed up unannounced and raided its data servers on Tuesday. The claim comes as Prime Minister Peter Magyar presses ahead with anti-corruption reforms following his landslide election victory over Viktor Orban in April.
Experts call NZ First's $1 billion oil and gas policy 'high-risk'
Brussels prepares first digital rulebook fine against Google, risking Trump's ire
Colombia's Petro calls for 'general strike' if successor repeals his reforms
