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

Senator Mark Warner said the NSA confirmed that Mythos Preview identified vulnerabilities in almost every classified system within hours during a controlled exercise.
US officials have clarified that Mythos discovered the flaws quickly rather than exploiting them, but there are still serious concerns about this capability.
Anthropic declined public release, sharing only with selected companies; Mozilla and others have validated its power, with thousands of critical bugs discovered in a matter of weeks

We now have another witness claiming that Mythos Preview is capable of quickly penetrating protected systems and this one is none other than a high-ranking member of the US government.

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

It is worth mentioning here that the break-in was controlled as it was part of an exercise carried out by the Anthropic team and the intelligence agency.

How powerful is Myth?

The Associated Press dug deeper and was told by an unidentified U.S. official that Mythos simply discovered vulnerabilities within hours, without necessarily exploiting them. Nonetheless, identifying a vulnerability that can theoretically be exploited for attacks against protected U.S. government systems should itself be cause for concern.

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

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

Mozilla said that with the help of Mythos, it was able to release more than 400 Firefox security bugs in April alone.

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

“Several have told us that their bug detection rate has increased more than tenfold,” the company said. “For example, Cloudflare found 2,000 bugs (400 of which were high or critical severity) on its critical path systems, with a false positive rate that the Cloudflare team considers better than that of human testers.

Via PK Press Club