IMF: AI could amplify cyberattacks, threaten global finance

The International Monetary Fund warned Thursday that artificial intelligence is increasing the risk of cyberattacks on the global financial system. In a blog post, IMF economists cited the controlled release of Anthropic’s Claude Mythos Preview and noted the model identified vulnerabilities across major operating systems and web browsers that could be exploited by unskilled actors.

The fund noted the combination of advanced AI tools and heavy reliance on a small number of cloud providers, widely used software platforms and a few AI models could allow a single breach to spread through linked banks, payment systems and markets.

The IMF explained a broad breach could move cyber incidents beyond operational outages into macro-financial shocks by triggering confidence losses, liquidity strains and forced asset sales that push prices down. The blog included this warning: “This foreshadows how fast-moving, AI-driven cyber risks could destabilize the financial system if not managed carefully.”

The IMF urged policymakers to treat cybersecurity as a core financial stability issue. It recommended stronger resilience standards for firms, more intensive systemic supervision, and enhanced cross-border coordination so breaches can be contained before they spread.

The fund also recommended defenders deploy AI-assisted tools that can operate at machine speed to detect threats, prevent fraud and speed incident response. The IMF noted many financial institutions are already using AI-based systems to improve detection and response times.

The analysis highlighted that emerging-market economies may face disproportionate exposure because they often have fewer resources for cyber defense and less regulatory capacity. The fund pointed to the geopolitical dimension, saying inconsistent oversight across jurisdictions could weaken the global system’s ability to withstand coordinated attacks.

Anthropic CEO Dario Amodei warned in a discussion with JPMorgan Chase CEO Jamie Dimon that there may be a six- to 12-month window to fix tens of thousands of software flaws identified by AI before risk levels rise. The IMF did not predict specific outcomes and called for immediate policy work to set standards and coordination mechanisms that reflect the speed at which AI can surface and exploit software vulnerabilities.