Powerful AI should not be released without safeguards, National Cyber Security Centre says

Last week, the company Anthropic — whose AI tools include the Claude platform — launched its new Mythos model but stopped short of a broad release, citing concerns it could expose previously unknown cybersecurity vulnerabilities. File picture

Powerful artificial intelligence tools with “potentially dangerous capabilities” should not be released without appropriate safeguards and meaningful engagement with national agencies, Ireland’s cyber security body has warned.

In a statement, the National Cyber Security Centre (NCSC) said the decision from AI firm Anthropic not to release its latest model on a general basis given its capabilities was a “responsible approach”.

“[We] would encourage all frontier AI developers to adopt similar practices widening the availability to trusted global industry and Government cyber security partners,” it said.

Last week, the company Anthropic — whose AI tools include the Claude platform — launched its new Mythos model but stopped short of a broad release, citing concerns it could expose previously unknown cybersecurity vulnerabilities.

The company has said the model is capable of identifying and exploiting weaknesses across "every major operating system and every major web browser”.

Anthropic said it would instead work with major companies such as Google, Apple, Amazon and Microsoft to use the tool to shore up the defences on its products and platforms.

What's your view on this issue?

You can tell us here

The announcement prompted action on both sides of the Atlantic, Trump administration officials convening an urgent meeting with bank CEOs to discuss the cyber risks posed by the Anthropic model, amid ongoing discussions between the US government and Anthropic.

Meanwhile, the Financial Times reported that British financial regulators were also holding urgent talks with the banks and the UK Government’s cyber security agency to discuss potential vulnerabilities in critical IT systems.

National Cyber Security Centre assessment

At home, the NCSC has reviewed the published technical material on Mythos and said the capabilities described “appear to represent a significant change in the way hardware and software vulnerabilities are identified and patched”.

“At present the advantage is with cyber defenders,” it said. “These new capabilities are in the hands of major technology vendors who are using them to identify and patch vulnerabilities in the hardware and software that underpins critical infrastructure worldwide.

“There is no indication that a comparable autonomous vulnerability discovery capability is available to threat actors at this time.” 

It added it will monitor developments and update national guidance as the situation develops.

It comes as senior officials warned TDs and senators that when it comes to AI “we’re in a race, whether we choose to accept it or not”.

Speaking at the Oireachtas AI Committee, NCSC director Richard Browne said: “The technical frontier is leaping ahead every week, and the role of managing cyber-related risks to society is becoming far more dynamic than we might like."