Vulnerabilities exposed in secret US government systems and software during testing by Anthropic Mythos

Senator Mark Warner testified that the NSA confirmed that Mythos Preview identified vulnerabilities in nearly all classified systems within hours during a controlled exercise
US officials clarified that Mythos found flaws quickly rather than exploiting them, but the capability remains a major concern
Anthropically withheld public release, sharing only with selected companies; Mozilla and others validated its strength, with thousands of critical bugs exposed within weeks

We now have another witness claiming that Mythos Preview is capable of quickly breaking into protected systems, and this is none other than a senior member of the US government.

According to the Associated Press, Senator Mark Warner of Virginia testified before a congressional hearing this month and said he was informed by National Security Agency (NSA) chief Joshua Rudd that Mythos “broke into almost all of our classified systems, not in weeks, but in hours.”

It is worth mentioning here that the intrusion was controlled as it was part of an exercise conducted by the anthropic team and the intelligence agency.

How powerful is the Mythos?

The Associated Press dug deeper and was informed by an unidentified US official that Mythos merely found vulnerabilities within hours, not necessarily exploiting them. Still, identifying a vulnerability that could theoretically be exploited to attack protected US government systems should in itself be cause for concern.

Mythos is an advanced AI model built by Anthropic, which was first introduced in early April this year. However, the company decided not to share it with the public because it was apparently too capable of detecting and exploiting software vulnerabilities.

Instead, Anthropic shared it with a handful of large companies to help them secure their systems before cybercriminals can use the tool. Since then, several companies have come forward to confirm Mythos’ strength, including Mozilla, which said the tool was “as capable” as the world’s top security researchers.

Mozilla said that using Mythos, it was able to patch more than 400 Firefox security bugs in April alone.

A month later, Anthropic said the 50 companies using the tool discovered more than 10,000 high-level critical security vulnerabilities in about two months.

“Several have told us that their speed of troubleshooting has increased by more than a factor of ten,” the company said. “For example, Cloudflare has found 2,000 bugs (of which 400 are of high or critical severity) across their critical path systems, with a false positive rate that Cloudflare’s team considers better than human testers.”

Via Pakinomist