Why Anthropic Is Limiting Access to Its Most Powerful AI, Claude Mythos

Anthropic has unveiled its advanced AI model, Claude Mythos Preview, but is restricting access as part of a cybersecurity initiative called Project Glasswing. The model is highly capable of identifying software vulnerabilities, raising concerns that it could be misused by hackers if widely released. To mitigate risks, Anthropic is rolling it out only to select partners, including Apple, Google, Microsoft, Nvidia, and Amazon Web Services, along with cybersecurity firms like CrowdStrike and Palo Alto Networks, to strengthen defensive security efforts.

Anthropic says its decision to limit access to Claude Mythos Preview followed intense internal debate around safety and impact. "There was a lot of internal deliberation," Dianne Penn, Anthropic's head of research product management, told CNBC in an interview. "We really do view this as a first step for giving a lot of cyber defenders a head start on a topic that will be increasingly important." The cautious rollout comes shortly after details of the model surfaced in a public data cache, raising concerns about its powerful cyber capabilities.

ALSO SEE: Apple iPhone 17e Review: Apple Hits A Sweet Spot It Has Been Chasing

The announcement also had ripple effects across the market, with cybersecurity stocks reacting to reports about the model's potential risks. At the same time, Anthropic is positioning the move as a long-term bet on safer AI deployment. "The dangers of getting this wrong are obvious, but if we get it right, there is a real opportunity to create a fundamentally more secure internet and world than we had before the advent of AI-powered cyber capabilities," CEO Dario Amodei wrote in a post on X.

Founded in 2021 by former OpenAI researchers, Anthropic has built its identity around responsible AI development. The launch of Project Glasswing comes amid ongoing discussions with U.S. government agencies and follows recent scrutiny over its stance on AI safety. The company says the model can uncover critical and previously undetected vulnerabilities, including a decades-old bug in OpenBSD, highlighting both its potential and the risks it carries.

Despite its capabilities, Anthropic has no immediate plans to release the model publicly. Instead, it is working with select partners to test its use in securing critical systems while avoiding misuse. "Cybersecurity is just going to be an area where this broad increase in capabilities has potential for risk, and thus we have to keep a really close eye on what's going on there," said Newton Cheng, reinforcing the company's cautious approach to deploying such powerful AI tools.

ALSO SEE: OnePlus Nord 6 vs Nord 5 Price Gap Explained: What Extra Do You Get For Rs 9,000?