Ask Onix
AI model sparks financial system security fears
Finance ministers, central bankers, and banking executives have voiced urgent concerns about an advanced AI system capable of exposing critical vulnerabilities in global financial infrastructure.
Crisis meetings at IMF summit
Canadian Finance Minister François-Philippe Champagne revealed that the Claude Mythos model, developed by Anthropic, dominated discussions at this week's International Monetary Fund meeting in Washington, D.C.
"The difference is that the Strait of Hormuz-we know where it is and how large it is. The issue with Anthropic is the unknown, unknown."
François-Philippe Champagne, Canadian Finance Minister
Champagne emphasized the need for robust safeguards to protect financial system resilience, calling the situation unprecedented in its uncertainty.
Model's capabilities under scrutiny
While Mythos has demonstrated an ability to identify security flaws in major operating systems, experts remain divided on its true threat level. The UK's AI Security Institute conducted the only independent assessment, noting its proficiency in exploiting weakly defended systems-but found it only marginally more advanced than Anthropic's previous model, Opus 4.
"Our testing shows Mythos Preview can exploit systems with weak security posture, and it's likely more models with these capabilities will emerge."
UK AI Security Institute report
Banks and regulators rush to respond
Top financial institutions, including Barclays, have been granted early access to test their systems against Mythos. Barclays CEO CS Venkatakrishnan warned of a "new world" of interconnected risks, stressing the urgency of understanding and patching vulnerabilities.
Bank of England Governor Andrew Bailey echoed these concerns, stating the central bank is "looking very carefully" at how AI advancements could escalate cybercrime risks by making it easier to detect and exploit IT weaknesses.
Governments and industry brace for impact
The U.S. Treasury has urged major banks to preemptively test their defenses before Mythos's public release. Meanwhile, industry sources suggest a rival U.S. AI firm may soon unveil a similarly powerful model with fewer safeguards.
James Wise, chair of the UK's £500 million Sovereign AI fund, described Mythos as the first of many AI systems capable of exposing vulnerabilities. His venture capital unit is investing in British AI security startups to develop tools that both identify and fix such flaws.
"We hope the models that expose vulnerabilities are also the ones that will fix them."
James Wise, Balderton Capital
What's next
Regulators and banks are accelerating efforts to assess Mythos's risks, with further testing and policy discussions expected in the coming weeks. The debate highlights broader tensions between AI innovation and financial stability.
