Anthropic’s most succesful AI mannequin has already discovered hundreds of AI cybersecurity vulnerabilities throughout each main working system and net browser. The corporate’s response was to not launch it, however to quietly hand it to the organisations chargeable for retaining the web working.
That mannequin is Claude Mythos Preview, and the initiative is named Project Glasswing.
The launch companions embody Amazon Net Companies, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Basis, Microsoft, Nvidia, and Palo Alto Networks.
Past that core group, Anthropic has prolonged entry to over 40 further organisations that construct or preserve vital software program infrastructure. Anthropic is committing as much as US$100 million in utilization credit for Mythos Preview throughout the trouble, together with US$4 million in direct donations to open-source safety organisations.
A mannequin that outgrew its personal benchmarks
Mythos Preview was not particularly skilled for cybersecurity work. Anthropic stated the capabilities “emerged as a downstream consequence of normal enhancements in code, reasoning, and autonomy”, and that the identical enhancements making the mannequin higher at patching vulnerabilities additionally make it higher at exploiting them.
That final half issues. Mythos Preview has improved to the extent that it principally saturates current safety benchmarks, forcing Anthropic to shift its focus to novel real-world duties–particularly, zero-day vulnerabilities. These flaws have been beforehand unknown to the software program’s builders.
Among the many findings: a 27-year-old bug in OpenBSD, an working system recognized for its sturdy safety posture. In one other case, the mannequin totally autonomously recognized and exploited a 17-year-old distant code execution vulnerability in FreeBSD–CVE-2026-4747–that permits an unauthenticated person wherever on the web to acquire full management of a server working NFS. No human was concerned within the discovery or exploitation after the preliminary immediate to search out the bug.
Nicholas Carlini from Anthropic’s analysis group described the mannequin’s potential to chain collectively vulnerabilities: “This mannequin can create exploits out of three, 4, or typically 5 vulnerabilities that in sequence provide you with some form of very refined finish consequence. I’ve discovered extra bugs within the final couple of weeks than I discovered in the remainder of my life mixed.”
Why is it not being launched?
“We don’t plan to make Claude Mythos Preview typically accessible resulting from its cybersecurity capabilities,” Newton Cheng, Frontier Crimson Group Cyber Lead at Anthropic, stated. “Given the speed of AI progress, it won’t be lengthy earlier than such capabilities proliferate, probably past actors who’re dedicated to deploying them safely. The fallout–for economies, public security, and nationwide safety–could possibly be extreme.”
This isn’t hypothetical. Anthropic had beforehand disclosed what it described as the primary documented case of a cyberattack largely executed by AI–a Chinese language state-sponsored group that used AI brokers to autonomously infiltrate roughly 30 world targets, with AI dealing with the vast majority of tactical operations independently.
The corporate has additionally privately briefed senior US authorities officers on Mythos Preview’s full capabilities. The intelligence group is now actively weighing how the mannequin might reshape each offensive and defensive hacking operations.
The open-source drawback
One dimension of Venture Glasswing that goes past the headline coalition: open-source software program. Jim Zemlin, CEO of the Linux Basis, put it plainly: “Prior to now, safety experience has been a luxurious reserved for organisations with giant safety groups. Open-source maintainers, whose software program underpins a lot of the world’s vital infrastructure, have traditionally been left to determine safety on their very own.”
Anthropic has donated US$2.5 million to Alpha-Omega and OpenSSF by way of the Linux Basis, and US$1.5 million to the Apache Software program Basis–giving maintainers of vital open-source codebases entry to AI cybersecurity vulnerability scanning at a scale that was beforehand out of attain.
What comes subsequent
Anthropic says its eventual objective is to deploy Mythos-class fashions at scale, however solely when new safeguards are in place. The corporate plans to launch new safeguards with an upcoming Claude Opus mannequin first, permitting it to refine them with a mannequin that doesn’t pose the identical degree of danger as Mythos Preview.
The aggressive image is already shifting round it. When OpenAI launched GPT-5.3-Codex in February, the corporate referred to as it the primary mannequin it had categorized as high-capability for cybersecurity duties underneath its Preparedness Framework. Anthropic’s transfer with Glasswing indicators that the frontier labs see managed deployment–not open launch–because the rising commonplace for fashions at this functionality degree.
Whether or not that commonplace holds as these capabilities unfold additional is, at this level, an open query that no single initiative can reply.
See Additionally: Anthropic’s refusal to arm AI is precisely why the UK desires it
Need to be taught extra about AI and large knowledge from business leaders? Try AI & Big Data Expo happening in Amsterdam, California, and London. The excellent occasion is a part of TechEx and is co-located with different main expertise occasions together with the Cyber Security & Cloud Expo. Click on here for extra data.
AI Information is powered by TechForge Media. Discover different upcoming enterprise expertise occasions and webinars here.
