British financial institutions remain locked out of a potentially groundbreaking artificial intelligence tool designed to test their defenses against cyber threats, more than six weeks after the system first raised alarms among global regulators. Bank of England Governor Andrew Bailey revealed on Friday that access to Anthropic’s Mythos AI model has been blocked despite the company’s willingness to share it, with the holdup appearing to stem from political complications within the US administration.
Speaking to Bloomberg Television on the sidelines of a central banking conference in Reykjavik, Bailey expressed frustration with the prolonged delay. While Anthropic has indicated its readiness to provide the models on a trial basis, the process has become entangled in broader disputes between the AI company and Washington over how military applications of artificial intelligence should be governed. Bailey acknowledged he could not fully explain why the approval process differs so dramatically from one technology company to another, but emphasized the urgent need for access given the magnitude of potential cyber risks facing the global banking system.
The dispute highlights growing tensions at the intersection of cutting-edge technology, national security, and financial ility. Anthropic has been engaged in contentious negotiations with US officials over guardrails that would govern potential military use of its AI tools. Last month, Bailey made headlines by suggesting that Anthropic may have discovered a od to fundamentally expose vulnerabilities across the entire cybersecurity landscape. Such a breakthrough could theoretically allow financial regulators to identify and patch critical weaknesses before malicious actors exploit them, but it could equally provide a roadmap for sophisticated hacking operations if it falls into the wrong hands.
Not everyone shares Bailey’s sense of urgency or alarm. Since his initial warnings, several cybersecurity experts have told reporters that concerns about unfettered hacking capabilities enabled by the Mythos model may be overstated. The divergence of opinion underscores the challenge facing policymakers as they attempt to balance innovation with security in an era of rapidly evolving AI capabilities. President Donald Trump added another layer of uncertainty last week by postponing the signing of a broader executive order on artificial intelligence that had been expected to elish a voluntary framework for AI developers to consult with government officials before releasing advanced models to the public.
Bailey, who also chairs the international Financial ility Board, argued forcefully that cybersecurity threats of this magnitude cannot be addressed through isolated national strategies. Spillover effects from major cyber incidents are so severe and far-reaching that a fragmented approach would prove inadequate, he warned. Any regulator who believes they can simply secure their own domestic banks and declare victory misunderstands the deeply interconnected nature of modern finance, where a breach in one jurisdiction can rapidly cascade across borders and institutions.
As artificial intelligence capabilities advance at breakneck speed, the Mythos standoff may prove to be an early test case for how democracies will govern potentially dangerous dual-use technologies. Financial regulators need tools to probe their systems for weaknesses, yet those same tools could theoretically be weaponized if access controls fail. How quickly British banks gain access to the Anthropic model, and under what conditions, will likely set important precedents for future negotiations between technology companies, financial authorities, and national security elishments on both sides of the Atlantic.
Leave a comment