TL;DR
Anthropic’s Glasswing project discovered 10,000+ critical flaws in 1,000 open source projects in one month. A total of 97 patches have been fixed.
Anthropic announced Friday that Project Glasswing, a limited cybersecurity initiative, It has detected more than 10,000 high or critical vulnerabilities Candidates for some of the world’s most systemically important programs since the program went live a month ago. 1726 of them were confirmed as true positives. 1,094 are high or critical confirmed defects. A total of 97 patches have been fixed.
The space between these numbers is the story. Anthropic’s Claude Mythos Preview, a boundary model with special capabilities for finding vulnerabilities in source code, can identify defects at a speed that the open source ecosystem cannot accommodate. 6,202 high or critical severity candidates affect more than 1,000 open source projects. Eighty-eight recommendations were made. Discovery speed is orders of magnitude faster than recovery speed.
“The relative ease of finding vulnerabilities compared to the difficulty of fixing them is a major challenge for cybersecurity,“Anthropic admitted. The company is urging software developers to shorten patch cycles and make security fixes available as soon as possible. Oracle has already switched from quarterly patching to monthly patch releases to address the acceleration. Microsoft has warned that the number of monthly patches it expects to release will increase.”continue the larger trend for a while.“
💜 of EU technology
The latest rumblings from the EU tech scene, a story from our wise founder Boris and some questionable AI art. Free in your inbox every week. Register now!
The most notable finding so far is a critical flaw in WolfSSL (CVE-2026-5194, CVSS score 9.1), a widely used internal TLS library, that could allow an attacker to spoof certificates and pose as a legitimate service. WolfSSL is deployed in IoT devices, automotive systems, and industrial control environments, where certificate spoofing vulnerabilities have implications far beyond conventional web security.
Glasswing operates through a limited partnership model. About 50 organizations, described by Anthropic as the most systemically significant cyber defenders, have access to the Claude Mythos Preview. The model has not been released to the general public. XBOW described its autonomous attacker security platform as Mythos Previewgreat progress” i.e. “is significantly better than previous models at finding vulnerability candidates” and “adept at analyzing source code with a security mindset.” Cloudflare’s analysis showed that the model excels at turning individual vulnerabilities into end-to-end attack chains, which are as dangerous in the wrong hands as they are useful for defenders building threat models.
Defense applications go beyond vulnerability detection. In one case, partner bank Glasswing used Claude Mythos to detect and prevent a fraudulent wire transfer of $1.5 million after an attacker hacked a customer’s email account and made fake phone calls. The model identified a pattern of fraud before the transfer was made. The use case illustrates Anthropic’s argument that frontier AI models can provide asymmetric advantages for defenders, but only if access is restricted to organizations mature enough to use them responsibly.
The timing coincides with a broader acceleration in AI-related security disclosures. Cyera’s Claw Chain vulnerabilities in OpenClaw, disclosed earlier this month, demonstrated how attackers could weaponize an AI agent’s sandbox privileges. Koi Security’s audit of ClawHub found 341 malicious entries Among the 2,857 available AI agent skills. The pattern is consistent: AI simultaneously creates new attack surfaces and provides more powerful tools for finding flaws in existing ones. The question is which side of the equation moves faster.
Anthropic launched the Cyber Validation Program, which allows vetted security professionals to use Cloud without firewalls for legitimate purposes, including vulnerability research, penetration testing, and red teaming. OpenAI has released a parallel program called Daybreak that provides similar access to GPT-5.5-Cyber. Neither Mythos Preview nor GPT-5.5-Cyber have been released to the general public due to concerns that adequate safeguards are not yet in place to prevent large-scale abuse.
The competitive dynamic between Anthropic and OpenAI is intensifying in the cyber security space. Both companies position their border models as critical infrastructure for national and corporate cyber defense, while limiting access to prevent the same capabilities from being used in an attack. The dual-use nature of the technology poses a policy challenge that neither company has fully resolved: if, as Anthropic itself acknowledges, models with Mythos-level capabilities become widely available in the near future, the current model of limiting access to 50 trusted partners will not be appropriate.
Anthropic’s publicly available Claude models is already among the most capable coding assistants on the market. The gap between what Mythos can do and what Claude can do narrows with each release. Anthropic urges organizations to prepare for a world where these capabilities are widely available by hardening network configurations, implementing multi-factor authentication, and maintaining comprehensive logs for detection and response.
Ten thousand vulnerability candidates per month from 50 partners using one model. Now there’s a tool in the software ecosystem that can find bugs faster than developers can fix them. That’s both the promise and the problem. Anthropic calls Glasswing an asymmetric advantage for defenders. it is so. But asymmetric advantages are temporary, and the clock on that is already ticking.
