US tech startup Anthropic says its new artificial-intelligence model, Claude Mythos, has helped find more than 10,000 software vulnerabilities.
Anthropic announced last month that about 50 partners, including major IT companies and financial institutions, would try Claude Mythos as part of a cybersecurity initiative.
It says most of them have each found hundreds of critical vulnerabilities in software, amounting to more than 10,000 in total.
Anthropic pointed out that human capacity to fix the flaws in software found by AI now creates a bottleneck.
The startup said it will work with critical partners, including US and allied governments, to expand its collaborative cybersecurity effort.
Anthropic decided against a general rollout for Claude Mythos out of concern it could be used in cyberattacks.
But the company has provided the AI model to a limited group of businesses and organizations.