Claude Mythos: When Code Proficiency Becomes a Security Weapon
Anthropic has built an AI model with Claude Mythos that identifies vulnerabilities in critical infrastructure. The model was not trained as a security tool. What this means for the cloud industry.
Anthropic has built a new AI model, Claude Mythos, capable of uncovering vulnerabilities in critical infrastructure–bugs that had remained hidden in code for 27 years and were missed by millions of automated tests. The model wasn’t trained as a security tool. It’s a coding model whose security capabilities emerged as a side effect. Instead of releasing it publicly, Anthropic is giving defenders a head start. Here’s what this means for the cloud industry.
Key Takeaways
- Claude Mythos achieves 93.9 percent on SWE-bench Verified–compared to 80.8 percent for Opus 4.6 (Anthropic, April 2026).
- On the CyberGym benchmark for vulnerability detection, Mythos scores 83.1 percent. Its security capabilities are a byproduct of superior coding proficiency.
- The model discovered a 27-year-old bug in OpenBSD and a 16-year-old bug in FFmpeg that 5 million automated tests had overlooked.
- Project Glasswing grants controlled access to over 40 organizations (AWS, Google, Microsoft, Apple, NVIDIA, CrowdStrike)–a defender-first approach.
- Committed around 87 million euros in usage credits and around 3.5 million euros for open-source security. All findings will be published within 90 days.
A Coding Model Becomes the Most Powerful Security Scanner
Claude Mythos wasn’t trained to be a hacker. Anthropic optimized the model to understand and write code better than any other available system. On SWE-bench Verified–the industry benchmark for bug-fixing capability–Mythos achieves 93.9 percent. Opus 4.6, the previously strongest public model, scores 80.8 percent.
The jump from 80 to 94 percent sounds like optimization. In practice, it’s a categorical shift. At these levels, the model solves nearly every real-world software task it’s given.
The real story lies in a side effect: anyone who understands code at this level also understands where code breaks. On the CyberGym benchmark for vulnerability detection, Mythos reaches 83.1 percent–compared to 66.6 percent for Opus. Anthropic likens this to a locksmith so skilled they can open any lock, despite never having been a burglar.
What Mythos Has Discovered in Practice
Three findings illustrate its reach.
In OpenBSD, Mythos uncovered a vulnerability in the TCP stack’s SACK implementation (Selective Acknowledgment–a mechanism for handling packet loss). 27 years old. The flaw enables remote denial-of-service attacks against any OpenBSD server. OpenBSD is considered one of the most secure operating systems. Its code is regularly audited by experienced security researchers. Still, the bug remained undetected.
In FFmpeg–the multimedia library that powers video and audio processing across virtually every platform–Mythos found a bug in the H.264 codec. 16 years old. Five million automated tests had passed through the affected code without triggering the vulnerability. FFmpeg handles the majority of video traffic on the internet. The project confirmed and patched the bug after Mythos flagged it.
In addition, Mythos identified several privilege-escalation vulnerabilities in the Linux kernel, including one in the DRR scheduler (Deficit Round Robin–an algorithm for distributing network bandwidth). An unprivileged user could gain full root access through this path.
The Market Shift: Coding Competence Becomes Security Competence
This is the point where this announcement stops being just an Anthropic press release and starts affecting the entire cloud industry.
If coding proficiency automatically generates security proficiency, then every next generation of coding models–whether from Anthropic, OpenAI, Google, or open source–will develop similar capabilities. This isn’t a feature that can be toggled on. It’s an emergent property that grows alongside code quality.
For cloud infrastructure, this means three things.
First, major providers will enhance their security scans with this technology. AWS, Azure, and GCP are Glasswing partners. Patches for previously unknown infrastructure bugs will roll out in the coming months. Cloud teams will benefit automatically.
Second, the bar for in-house code is rising. If an AI model can detect 27-year-old bugs missed by manual audits, saying “we ran a penetration test” will no longer suffice as proof of security. CI/CD pipelines will need to integrate AI-powered security scans as a standard gate.
Third, a new category of security tools is emerging. The market for AI-driven vulnerability detection will fundamentally shift within 12 to 24 months–not because Mythos is a specialized security tool, but because it proves that any strong coding model inherently becomes a strong security model.
Project Glasswing: Controlled Distribution to the Defenders
Anthropic has opted against a public release. Instead, Mythos operates under the name Project Glasswing in a controlled environment. The partner list includes: AWS, Apple, Google, Microsoft, NVIDIA, Cisco, CrowdStrike, JPMorgan Chase, Palo Alto Networks, Broadcom, the Linux Foundation, and over 30 additional organizations.
The logic: If a model can find vulnerabilities in the infrastructure underpinning the internet, the operators of that infrastructure should get access first. Three concrete commitments: All findings will be publicly shared within 90 days. around 87 million euros in usage credits are available to partners. around 3.5 million euros will go directly to open-source security groups.
Patches are already rolling out. FFmpeg has confirmed and fixed the 16-year-old bug. Fixes for OpenBSD and the Linux kernel are now being deployed.
What Cloud Teams Should Do Now
In the short term, there’s little to do. If you have automatic updates enabled, you’re largely protected. Glasswing patches are delivered through standard update channels from cloud providers and OS distributions.
In the medium term, engineering teams should evaluate AI-powered code analysis tools for their tech stack. Not because Mythos will be available tomorrow, but because the next generation of coding assistants will bring similar capabilities. Teams that prepare their CI/CD pipelines for AI-driven security scans today will gain a competitive edge.
In the long term, the balance of power is shifting. Defenders are gaining tools that were previously available only to elite attackers. But the same technology will become accessible to attackers once comparable open-source models emerge. Security remains an arms race. The difference this time: defenders now have a head start.
Frequently Asked Questions
What is Claude Mythos and how does it differ from Claude Opus?
Claude Mythos is Anthropic’s next-generation model, achieving 93.9 percent on SWE-bench Verified–compared to 80.8 percent for Opus. The security capabilities aren’t a trained feature but a byproduct of superior coding proficiency.
Why isn’t Anthropic releasing the model publicly?
A model capable of identifying vulnerabilities in critical infrastructure and combining them into exploit chains would become an attack tool in the wrong hands. Through Project Glasswing, defenders gain first access so bugs can be patched before becoming public knowledge.
Do smaller businesses also benefit from Project Glasswing?
Indirectly, yes. Patches are rolled out via regular updates from major platforms. Users of AWS, Azure, or GCP benefit automatically. Direct access is currently limited to over 40 partner organizations.
Will other AI labs develop similar models?
The Mythos results show that security capabilities are a side effect of advanced coding skills. Every upcoming frontier coding model will likely develop similar abilities. Whether OpenAI, Google, and Meta will adopt the same controlled approach remains to be seen.
What should DevOps teams do right now?
Keep all systems up to date–the Glasswing patches arrive through regular updates. Evaluate AI-powered security scans for your CI/CD pipeline. Raise your security standards: if AI can uncover 27-year-old bugs, an annual pentest is no longer sufficient.
Editor’s Reading Picks
More from the MBF Media Network
- → Deepfake Attacks on the C-Suite: How AI Voices Steal Millions (SecurityToday)
Image credit: AI-generated conceptual image (FLUX.2) – not a product photo
Image source: AI-generated (May 2026), C2PA certificate embedded in image
Translated from the German original using artificial intelligence. The German version is authoritative.

