AI startup Anthropic said it created a model that is too powerful to be released to the public.
Mythos exposed thousands of software vulnerabilities in common applications for which no fix exists, according to Anthropic, which instead shared a version of the software with certain companies to bolster defenses against AI-powered hacking.
The situation is “almost reminiscent of Y2K,” except that in this case, cyber defenses are already “abysmal” and AI coding makes them more vulnerable, Semafor’s tech editor wrote. “It would be great if Mythos were the wakeup call the world needed to finally do something about cybersecurity, but that seems unlikely.”
Meanwhile, Meta released its latest AI model Wednesday, which lagged rivals on coding abilities.