Global financial leaders are sounding the alarm over Anthropic's Mythos AI, a powerful tool designed to test and potentially exploit cybersecurity vulnerabilities, with discussions echoing through an IMF meeting in Washington, DC.

Mythos is framed as a potentially transformative but uncertain threat that could change how financial systems are secured as AI capabilities accelerate.

Industry figures warn of existential risks to banks and payment networks, triggering regulatory and central-bank scrutiny worldwide.

Defensive measures are being taken, such as air-gapped databases and a stronger emphasis on human factors like social engineering.

Indian fintechs like Razorpay are pushing for Mythos access to test defenses, reflecting a race to bolster cybersecurity ahead of broader deployment.

India, with a large engineering workforce, is a major hub for Anthropic’s Claude model and a key market for developers using AI for app building and IT modernization.

Anthropic has restricted Mythos access and is expanding it gradually through Project Glasswing to a select set of partners for defensive use, including major tech and financial players.

The Glasswing program limits access to prevent security risks from wide exposure, focusing on entities like AWS, Microsoft, Google, and JPMorgan Chase.

The aim of Glasswing is to enable partners to scan networks and patch flaws before disclosure, rather than public deployment.

A detailed 245-page technical document on Mythos highlights self-correcting coding abilities and notes a notable performance uptick over the prior model on a major math competition benchmark.

A key data issue identified is fragmentation; the IMF urges treating data sharing as infrastructure with standardized APIs to enable cross-institution threat detection.

Independent testing by the UK AI Security Institute finds Mythos can exploit weak systems but is not dramatically more advanced than Opus 4, signaling caution rather than alarm.