Banks Test Systems After Anthropic Mythos Warning | Silicon UK

US Treasury seeks access to latest Claude model, Wall Street banks carry out tests after Anthropic warns of security risks

Getting your Trinity Audio player ready...

The US Treasury is seeking to gain access to Anthropic's Claude Mythos AI model to use it to hunt for vulnerabilities in its own systems, Bloomberg reported, following the start-up's alert about the model's capabilities.

Sam Corcos, the Treasury's chief information officer, is aiming to gain access to the model as soon as this week, according to the report.

Security testing

Anthropic said earlier this month that it was releasing Mythos to a limited number of institutions to conduct security testing.

It said during its own testing, it found that Mythos was capable of identifying and exploiting vulnerabilities "in every major operating system and every major web browser when directed by a user to do so", such as web browser exploit that chained together four vulnerabilities.

Mythos is the latest version of Anthropic's flagship AI model, Claude, the successor to Claude Opus 4.6.

The company's alert has renewed speculation around the misuse of powerful AI models, which have already been used to carry out hacks, steal data, and make fraudulent activities appear more plausible.

Implications

Wall Street banks have begun testing their systems with Mythos internally, while Bank of England governor Andrew Bailey said this week that regulators were examining the implications of the model.

Speaking at an event at Columbia University in New York City, Bailey said regulators wanted "to what extent is this new version of the product going to be able to, in a sense, identify vulnerabilities in other systems which can be exploited for cyberattack purposes".