FOLLOW US

Anthropic said Tuesday it is restricting access to a new artificial intelligence model capable of dramatically accelerating cyberattacks, opting instead to share it with a select group of major technology and cybersecurity companies to strengthen defenses before a broader rollout.

The company confirmed that firms including Amazon, Apple, Cisco, Google, JPMorgan Chase and Microsoft will be given early access to its latest system, known as Mythos, for defensive applications. Those uses include identifying software vulnerabilities and testing how products withstand potential hacking techniques.

The model — formally named “Claude Mythos Preview” — is not being released publicly due to concerns it could be misused by hackers or intelligence operatives. The decision reflects growing unease among policymakers in Washington and technology leaders in Silicon Valley about the rapid evolution of AI-driven cyber capabilities.

Security specialists told CNN that AI systems capable of autonomously scanning for weaknesses represent a major shift in the threat landscape. Compared with traditional methods, a single AI agent could detect and exploit vulnerabilities at a speed and scale far beyond what human hackers can achieve.

“We did not feel comfortable releasing this generally,” Logan Graham, who heads the team at Anthropic its AI models’ defenses, told CNN. “We think that there’s a long way to go to have the appropriate safeguards.”

Anthropic has also conducted briefings with senior officials across the US government on both the offensive and defensive potential of the model, according to a company representative. The firm has offered to assist authorities in evaluating and testing the technology further.

Executives at Anthropic say providing Mythos to companies that collectively serve billions of users could help close the gap between attackers and defenders. By identifying flaws early, the aim is to prevent major vulnerabilities in widely used software — such as web browsers and operating systems — from reaching the public.

Additional organizations granted access include semiconductor companies Broadcom and Nvidia, the nonprofit Linux Foundation, and cybersecurity firms CrowdStrike and Palo Alto Networks.

“If models are going to be this good — and probably much better than this — at all cybersecurity tasks, we need to prepare pretty fast,” Graham told CNN. “The world is very different now if these model capabilities are going to be in our lives.”

Details about Mythos surfaced last month in a leaked blog post that described the system as “far ahead” of competing models in cyber capabilities. The post warned that it “presages an upcoming wave of models that can exploit vulnerabilities in ways that far outpace the efforts of defenders,” according to earlier reporting by Fortune.

While some experts told CNN that fears about misuse may be overstated, they acknowledged the leak underscored a broader concern: without intervention, AI could widen the gap between those launching cyberattacks and those trying to stop them.

Anthropic says the model has already demonstrated significant results, claiming it has uncovered “thousands” of previously unknown software flaws in recent weeks — a pace the company says exceeds that of human researchers. CNN was not able to independently confirm the figure.

Such vulnerabilities are highly valuable, often sought by governments and cybercriminal groups because they can enable covert and sophisticated intrusions.

Even before Mythos, AI has been used in cybersecurity defense. In December, researcher Gadi Evron and others introduced a tool built on Anthropic’s Claude model designed to automatically generate fixes for critical software weaknesses.

“Unlike attackers, defenders don’t yet have AI capabilities accelerating them to the same degree,” Evron, the founder of AI security firm Knostic, told CNN. “However, the attack capabilities are available to attackers and defenders both, and defenders must use them if they’re to keep up.”