Powell, Bessent Warn Banks About Security Risks From Anthropic's Mythos AI: Bloomberg

U.S. Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell reportedly convened a gathering with Wall Street slope CEOs earlier this week to pass astir cybersecurity risks tied to a caller artificial quality exemplary from Anthropic.

According to a study by Bloomberg, the gathering included executives from Citigroup, Bank of America, Wells Fargo, Morgan Stanley, and Goldman Sachs. Officials discussed Anthropic’s caller AI exemplary Mythos, which has precocious drawn wide interest implicit its evident precocious cybersecurity capabilities.

Officials convened the gathering to guarantee banks recognize the risks posed by systems susceptible of identifying and exploiting bundle vulnerabilities crossed operating systems and web browsers, and to promote institutions to fortify defenses against imaginable AI-assisted cyberattacks targeting fiscal infrastructure.

Security researchers person warned that tools susceptible of automatically discovering vulnerabilities could accelerate some antiaircraft information enactment and malicious hacking if misused.

Anthropic’s Mythos exemplary archetypal surfaced online successful March aft draught materials astir the strategy leaked online, revealing what the institution described arsenic its astir susceptible AI exemplary yet. In testing, the strategy reportedly recovered thousands of antecedently chartless bundle vulnerabilities, including zero-day flaws crossed large operating systems and web browsers.

Anthropic researchers said successful a study earlier this week that Mythos Preview’s vulnerability-discovery capabilities were not intentionally trained, but alternatively emerged from broader improvements successful the model’s coding, reasoning, and autonomy.

“The aforesaid improvements that marque the exemplary substantially much effectual astatine patching vulnerabilities besides marque it substantially much effectual astatine exploiting them,” the steadfast wrote.

Because of those capabilities, Anthropic has restricted entree to a tiny radical of cybersecurity organizations.

Anthropic's Mythos Safety Report Shows It Can No Longer Fully Measure What It Built

“Given the spot of its capabilities, we’re being deliberate astir however we merchandise it,” Anthropic said successful a statement. “As is modular signifier crossed the industry, we’re moving with a tiny radical of aboriginal entree customers to trial the model. We see this exemplary a measurement alteration and the astir susceptible we’ve built to date.”

To code that risk, Anthropic is investigating Mythos done Project Glasswing, a collaboration with large exertion and cybersecurity companies that uses the exemplary to place and spot vulnerabilities successful captious bundle earlier attackers tin exploit them.