USA

Finance ministers and bankers raise serious concerns about Mythos AI model

Upgrade to get full access to all premium news on Yahoo Finance and more great articles like this free preview.

A Silver or Gold subscription plan is required to access premium news articles.

Finance ministers, central bankers and financiers have expressed serious concerns about a powerful new artificial intelligence model they fear could undermine the security of financial systems.

The development of the Claude Mythos model by Anthropic led to crisis meetings after it found vulnerabilities in many major operating systems.

Experts say it has a potentially unprecedented ability to identify and exploit cybersecurity weaknesses; but others warn that more testing is needed to get an accurate understanding of its abilities.

Canadian Finance Minister François-Philippe Champagne told the BBC that Mythos was discussed extensively at the International Monetary Fund (IMF) meeting in Washington DC this week.

“This situation is certainly serious enough to attract the attention of all finance ministers,” he said.

“The difference is that we know where the Strait of Hormuz is and we know how big it is… The problem we face with Anthropic is that it is the unknown, the unknown.”

“This requires a lot of attention, so we have security measures in place and we have processes in place to make sure we ensure the resilience of our financial systems,” he added.

Mythos is one of Anthropic’s latest models, developed as part of a broader AI system called Claude that rivals OpenAI’s ChatGPT and Google’s Gemini.

It was revealed by Anthropic earlier this month, when developers responsible for testing AI models and performance of so-called “misaligned” tasks that run counter to human values, goals and behaviors said the AI was “strikingly capable at computer security tasks.”

Anthropic did not release the model, citing concerns that it could reveal old software bugs or find ways to easily exploit system vulnerabilities.

Instead, it made Mythos available to tech giants like Amazon Web Services. CrowdStrikeAs part of a venture called Microsoft and Nvidia Glass Wing Project – calls it “an effort to secure the world’s most critical software.”

Mythos is part of Anthropic’s Claude system, the company’s family of AI models and AI assistant of the same name. [NurPhoto via Getty Images]

Anthropic on Thursday released a new version of the existing model Claude Opus, saying it will allow Mythos’ cyber capabilities to be tested on less powerful systems.

Concerns expressed about Mythos may surpass those around previous AI models, but some cybersecurity experts question how justified they are. – especially given that the model hasn’t been tested by the wider industry to see how capable it actually is.

The AI Security Institute in the UK was granted access to a preview version of this and has published the only independent report on the model’s cybersecurity capabilities.

Its researchers noted that it was a powerful tool that could find many vulnerabilities in vulnerable environments, but suggested that Mythos was not significantly better than Claude’s predecessor, Opus 4.

“Our testing shows that Mythos Preview can exploit systems with a weak security posture, and it is likely that more models with these capabilities will be developed,” the report’s authors said.

It’s also not the first time an AI developer has claimed that their model’s capabilities mean they shouldn’t be released. Critics suggest this is a excitement-mongering tactic.

OpenAI expressed similar fears in February 2019 when it chose to delay the release of GPT-2, an earlier version of its models that now powers its flagship tool, ChatGPT.

Senior bankers will be given access to the model in advance to test their systems.

CS Venkatakrishnan, chief executive of Barclays, told the BBC: “The situation is serious enough for people to be concerned.

“We need to understand this better and understand the vulnerabilities that are exposed and fix them quickly.”

“This is what the new world will be like,” he added, referring to a much more connected financial system with both opportunities and vulnerabilities.

Developer Anthropic said the model has already revealed numerous vulnerabilities in some critical operating systems, financial systems and web browsers, while governments and banks are being offered access to help protect their own systems before it is made public.

Bank of England governor Andrew Bailey told the BBC that this development should be taken very seriously: “We now need to look very carefully at what this latest AI development could mean in terms of cybercrime risk.”

He added: “As a result, there may be an improvement in artificial intelligence and modeling that makes it easier to detect existing vulnerabilities in some kind of underlying IT system and then obviously cybercriminals – bad actors – can try to exploit them.”

The US Treasury confirmed that Anthropic raised the issue ahead of Mythos’ public release by encouraging major banks to test their systems.

A leading US AI company may soon launch a similarly powerful model but without the same security measures, financial industry sources said.

James Wise, partner at Balderton Capital, is head of the Sovereign AI unit, a venture capital fund backed by £500m of government funding that will invest in British AI companies.

He said Mythos was “the first of more powerful models” that could reveal the weak points of systems.

He told the BBC’s Today Programme that his unit was “investing in British AI companies that are tackling this, namely companies working in AI safety and security.”

“We hope that the models that reveal vulnerabilities will also be the models that will fix them.”