[AI+Anthropic] U.S. CISA reportedly uses Mythos to review government code vulnerabilities

The U.S. Cybersecurity and Infrastructure Security Agency (CISA) is reportedly using Anthropic’s AI model Mythos to review government software, showing that even though there are ongoing disagreements between Anthropic and the White House, the government is still eager to adopt Anthropic’s tools.

Reuters, citing sources, said that CISA is using Mythos to scan government code repositories to look for vulnerabilities that could potentially give foreign spies and cybercriminals an opening. The sources said the scanning work is carried out by CISA’s Attack Surface Assessment team. The team is a CISA division responsible for conducting digital security assessments and hacker drills across the entire government.

The review is said to have found a large number of vulnerabilities, but it is not possible to determine how much government code the team has reviewed, nor the nature or severity of the vulnerabilities.

This year, Anthropic’s relationship with the government deteriorated after it refused to let the U.S. military use its AI model for domestic surveillance and fully autonomous weapon systems. The government subsequently placed Anthropic on a supply chain blacklist.

According to Axios, although Mythos has been blacklisted, the U.S. National Security Agency (NSA) began using it as early as April. Reports at the end of last month said that NSA analysts tested Mythos in a classified environment and were impressed by its capabilities.

However, after Anthropic rolled out a public version of Mythos called Fable, the White House suddenly demanded that the company ban foreigners from operating the model, causing the model to be suspended worldwide until it was lifted just last week.