Senator Mark Warner revealed that the NSA confirmed Mythos Preview successfully identified vulnerabilities across nearly all classified systems within mere hours during a controlled exercise.
U.S. officials clarified that while Mythos discovered flaws swiftly, it did not exploit them, yet the potential remains a significant concern.
Anthropic opted against a public release of Mythos, providing access solely to select corporations; validation from Mozilla and others corroborates its efficacy, uncovering thousands of critical bugs within weeks.
A notable revelation has emerged regarding Mythos Preview, now emphasized by a prominent U.S. government official.
As reported by the Associated Press, Senator Mark Warner of Virginia testified in a congressional inquiry this month, disclosing that NSA chief Joshua Rudd informed him that Mythos “broke into nearly all of our classified systems, not in weeks, but within hours.”
It is imperative to note that this breach was orchestrated in a controlled environment, part of an exercise conducted collaboratively by the Anthropic team and the intelligence agency.
Assessing the Power of Mythos
Further investigation by the Associated Press revealed insights from a confidential U.S. source, indicating that Mythos identified vulnerabilities expeditiously, rather than executing exploitation.
Nevertheless, the ability to pinpoint weaknesses capable of being exploited in protected U.S. government systems warrants substantial apprehension.
Mythos, an advanced artificial intelligence model developed by Anthropic, was initially unveiled in the early days of April. However, due to its alarming proficiency in uncovering and potentially exploiting software vulnerabilities, the firm opted against a broad public release.
Instead, Anthropic elected to grant access to a select group of major corporations to assist them in fortifying their networks before malicious entities could wield the tool.
Numerous companies have since validated Mythos’ capabilities, including Mozilla, which asserted the tool is “every bit as capable” as the leading security researchers in the field.
Mozilla reported that, with the assistance of Mythos, it was able to rectify over 400 Firefox security vulnerabilities in the month of April alone.
A month hence, Anthropic announced that the 50 companies utilizing the tool had collectively identified more than 10,000 critical and high-severity security vulnerabilities in a span of approximately two months.
“Several have communicated that their rate of bug discovery has surged by more than tenfold,” the company asserted.
“For example, Cloudflare reported finding 2,000 bugs, 400 of which possess high or critical severity, across their critical-path systems, demonstrating a false positive rate that has been evaluated as superior to that of human testers.”
Source link: Techradar.com.
