UK banks blocked from cyber AI tool Mythos get offer from rival OpenAI

OpenAI has offered 9 large UK banks entree to its cyber information AI instrumentality GPT-5.5 Cyber, arsenic its fierce rival Anthropic has blocked them successful previews of its version, Claude Mythos.

Both tools are designed to find hidden information weaknesses successful integer systems and tin outperform humans astatine immoderate hacking and cyber-security tasks.

Mythos caused a disturbance erstwhile it was announced successful April, with Anthropic claiming it had recovered a weakness successful 1 bequest strategy which had remained hidden for astir 30 years.

Some concern ministers, cardinal bankers and financiers person since expressed superior concerns astir it, fearing the exemplary could undermine the information of each fiscal systems.

The AI Security Institute, which has tested some Anthropic's Mythos and OpenAI's GPT-5.5 Cyber, said successful its study that they reached "a akin level of performance" successful the tasks it acceptable them.

Last week Bank of England politician Andrew Bailey warned that UK banks were inactive incapable to entree Mythos to trial the information of their ain integer systems and apps.

Former UK Chancellor George Osborne, who is present a elder enforcement astatine OpenAI, told the BBC that Bailey had not contacted him directly. He said his steadfast did not privation to "hide [5.5 Cyber] distant oregon support it to ourselves" but confirmed that, similar Mythos, it would not go disposable to all.

"The cardinal things with these tools is that they request to beryllium successful the hands of the close people," helium said.

"We privation to marque definite that the forces that are establishing bid successful our democracies person these tools, and the forces that privation to disrupt america oregon perpetrate crime, bash not."

The UK banks which volition present summation entree to OpenAI's GPT 5.5 Cyber see Lloyds Banking Group, HSBC and Nationwide. NatWest and Santander already person entree nether existing agreements.

The BBC has asked Anthropic whether it plans to assistance UK fiscal institutions' entree to Mythos.

Anthropic initially opened it up to a corporate of 42 companies, mostly different US tech firms.

OpenAI has been much generous with entree to Cyber 5.5, opening it up successful the EU and besides to banks successful Japan and Canada, among others.

Both firms complaint for usage though Anthropic enactment $100m towards Mythos previews.

"I don't needfully expect these tools to aboveground things humans wouldn't find yet but they are relentless and incredibly thorough successful sorting done the millions of lines of codification which are successful banking apps alone," said Prof Alan Woodward, a cybersecurity adept from Surrey University.

"In the UK, immoderate of the codification which banking systems inactive tally connected is incredibly old, and AI volition beryllium precise utile astatine uncovering problems there.

"They tin bash galore weeks' worthy of enactment successful minutes – but they bash besides aboveground mendacious positives, truthful you inactive request a quality successful the loop to cheque their findings."