Why Anthropic’s new AI device has central banks and authorities terrified: ALEX BRUMMER

The sombre standoff between the US and rest of the world over the festering economic damage caused by the Iran war has been the main source of angst at this week’s high level IMF gatherings.

Below the radar, a new scary economic, financial and national security threat has rapidly moved to the top of the global agenda.

San Francisco-based Anthropic revealed its engineers have developed a tool, Claude Mythos, so powerful that it cannot be publicly released.

The disclosure has US authorities, central bankers and finance ministers on alert.

Mythos’ creation is described by Anthropic as providing a ‘next generation’ capability for offensive cyber attacks that can infiltrate previously impenetrable software infrastructure and find hidden weaknesses.

Just as governments and corporations around the world are adjusting to the capabilities of AI, along comes a new technology – threatening to national security, official payments systems and corporate life.

On the attack: Anthropic may unsuspectingly have increased odds that Mythos falls into the hands of those able to penetrate robust cyber defences

Britain knows from the damaging cyber attacks on Marks & Spencer, the Co-op and Jaguar Land Rover in the last year the impact on financial performance.

Anthropic says Mythos found a flaw in code that had been tested five million times without detection.

It regards Mythos as so lethal that it is the first AI model ever to be restricted from users because of its destructive cybersecurity potential. In response to this capability, it chose a limited distribution of the tool to several tech firms and banks to pre-emptively test, identify and defend vulnerabilities at scale.

The consortium – named Project Glasswing – comprises Amazon, Apple, Google, Cisco, CrowdStrike, JPMorgan Chase, Microsoft and Nvidia.

There already is an ongoing legal dispute between Anthropic and the Pentagon over whether their existing access to military systems breaches ethical guidelines.

Despite this fracture, the US government is considering whether it should allow Claude Mythos access to critical systems.

The Bank of England, in common with other UK government agencies, finds itself under frequent attack from hackers.

It constantly upgrades its cyber defence capabilities and, so far, vital parts of its work, such as bank payments systems, have proved well protected. 

Yet if Claude Mythos is as lethal as Anthropic purports, it could in wrong hands be truly devastating to Britain’s sophisticated financial markets. 

London, after all, is the centre of foreign exchange derivative trades, with a daily turnover of $4.32 trillion alone.

The speed at which Anthropic engineers have come up with the latest tool rocked financial leaders. Regulators are being asked to grapple with an unknown technology which affects all financial institutions.

Among obvious worries is that Anthropic, by releasing Mythos to commercial players for testing, may unsuspectingly have increased odds that it falls into the hands of those able to penetrate robust cyber defences. 

The tool emerged so speedily that there has been no opportunity to build measures to mitigate the onslaught.

Anthropic is also creating a fresh divide between the US and Europe and, for that matter, the rest of the world. It is giving US financial groups, such as JP Morgan, the ability to fix its systems. That might cause bad players to turn their attention elsewhere, such as the City of London.

That is a terrifying prospect.