Anthropic says its new AI model is so good at spotting security risks, you can’t use it

AI developer Anthropic says it’s the latest Claude The AI ​​model is too good at finding cybersecurity vulnerabilities to be made public. Instead, the company provides the tool to large technology infrastructure providers so they can patch the flaws they discover.

In late March, Word started to leak out Anthropic’s latest model of artificial intelligence, It’s called the Claude Mythos (PDF), it will be a leap forward for the company’s AI technology. Now, the company has previewed its capabilities and warned that Mythos poses a significant cybersecurity threat, as it does Capabilities represent a leap forward In finding and exploiting online security vulnerabilities.

“AI models have reached a level of coding ability where they can outperform all but the most adept humans at finding and exploiting software vulnerabilities,” the company said. he said in a blog post Tuesday. Anthropic said Mythos Preview, which has not been released to the public, has already found what it says are thousands of critical vulnerabilities “in every major operating system and web browser.” When requested for comment, an Anthropic representative directed CNET to the company’s blog post.

To address cybersecurity risks, Anthropic said it would launch a consortium It’s called the Glasswing Project Which include Apple, Amazon Web Services, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, Linux Foundation, Microsoft, Nvidia, and Palo Alto Networks. Those organizations and more than 40 others will have access to Mythos to begin work on shoring up defenses against AI attacks and exploits, Anthropic said. It has allocated $100 million in usage credits to Mythos and $4 million in donations to open source security organizations.

“The risks of getting this wrong are clear, but if we get it right, there is a real opportunity to create an Internet and a world that is fundamentally safer than it was before the advent of AI-powered cyber capabilities,” Anthropic CEO Dario Amodei said. Published on X.

In a The video was posted on YouTube Around the Glasswing project, company leaders including Microsoft, the Linux Foundation, and Anthropic discussed the damage software vulnerabilities can cause.

Large cloud computing companies have already worked with the new model to find vulnerabilities. “What we found was helpful,” said Anthony Greco, Cisco’s chief security and trust officer. he wrote in a blog post. “Now the real work begins. AI-powered analysis reveals data at a scale and depth that legacy frameworks were not designed to accommodate.”

Amazon Web Services has already said the model Find ways to enhance your code Even in their most tested systems. Amy Herzog, vice president and chief information security officer at AWS, described the Claude Mythos Preview as “a step change in the thinking and capabilities of AI for cybersecurity.”

How important is this new model?

The phenomenon of artificial intelligence being able to discover and exploit vulnerabilities in software is not new DARPA’s Grand Cyber ​​Challenge Michel Salat, director of threat intelligence at the company, said that the company has witnessed several cases in which artificial intelligence has attracted attention in this field. Nortonantivirus provider.

But now there is AI technology that can be available to anyone with some of these capabilities. “Anthropic’s Glasswing project is focused on protecting this powerful technology, which could be transformative in vulnerability research but also poses a serious risk if misused for malicious purposes,” Salat said in an email. “Although it represents a huge step forward from the best current models such as Opus 4.6The core capability already exists today, and vulnerability research is quickly emerging as one of the primary real-world use cases for AI in cybersecurity.

National policymakers, who were going back and forth On the need for federal AI regulationThe consortium’s progress is likely to be closely watched.

Senator Mark Warner He praised the initiative in a statement. “I applaud these leading companies for recognizing this threat and proactively sharing information, capabilities and computing power to better protect our critical infrastructure,” the Virginia Democrat said. “As AI dramatically accelerates the discovery of new vulnerabilities, I hope that the industry will in turn accelerate the patching process and reprioritize it.”

Warner, whose state is a hotbed for AI data centers, recently called for a proposed moratorium on data center construction.”follyBut he also warned of the risks this poses to society The rapid development of artificial intelligence is leading to massive job losses.