Anthropic Unveils Powerful AI Vulnerability Hunter, Restricts Public Access

Tech giant partners with Anthropic to find zero-day flaws before bad actors can exploit them.

Apr. 8, 2026 at 2:37pm

Got story updates? Submit your updates here. ›

Anthropic has developed a powerful AI model called Claude Mythos Preview that can autonomously discover and exploit software vulnerabilities across major operating systems. To prevent these capabilities from falling into the wrong hands, Anthropic has launched Project Glasswing, a $100 million initiative that gives 12 major tech companies direct access to the AI to find and fix critical bugs before they can be exploited.

Why it matters

The ability of an AI to surpass even the most skilled human hackers at finding and exploiting vulnerabilities raises significant cybersecurity concerns. If this technology were to proliferate, it could lead to a surge in cyberattacks and data breaches. By partnering with leading tech firms, Anthropic hopes to get ahead of the threat and shore up defenses before the AI's capabilities become more widely available.

The details

Anthropic's Claude Mythos Preview AI has autonomously discovered thousands of vulnerabilities, including a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg that survived millions of tests. Unlike previous models, Mythos can write working exploits entirely on its own, posing a serious threat to even the most secure systems. Anthropic warns that similar AI capabilities could spread to competitors within 6-18 months, making 2026 attacks 'significantly more likely'.

  • Anthropic announced Project Glasswing on April 8, 2026.
  • Anthropic warns that similar AI capabilities could spread to competitors within 6-18 months.

The players

Anthropic

An artificial intelligence company that has developed the powerful Claude Mythos Preview AI model.

Newton Cheng

An Anthropic executive who stated that the Mythos model 'can surpass all but the most skilled humans at finding and exploiting software vulnerabilities'.

Project Glasswing

A $100 million initiative launched by Anthropic to partner with 12 major tech companies, including Amazon, Apple, Google, and Microsoft, to find and fix critical software vulnerabilities before they can be exploited.

Got photos? Submit your photos here. ›

What they’re saying

“This model 'can surpass all but the most skilled humans at finding and exploiting software vulnerabilities.'”

— Newton Cheng, Anthropic executive

What’s next

Anthropic has briefed government officials on the potential economic and national security fallout from the spread of similar AI capabilities, and is racing to get ahead of the threat by partnering with major tech companies through Project Glasswing.

The takeaway

Anthropic's development of a powerful AI vulnerability hunter capable of surpassing even skilled human hackers highlights the growing cybersecurity challenges posed by advanced AI technologies. By restricting public access and partnering with leading tech firms, Anthropic is attempting to get ahead of the threat and shore up defenses before these capabilities become more widely available.