😺 Too dangerous to release

2026-04-08 · Source: The Neuron · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Healthcare Systems & Policy · Depth: Fundamental Awareness, long

Summary

Anthropic has unveiled "Project Glasswing," a cybersecurity initiative centered on its unreleased Claude Mythos Preview model, which the company claims is exceptionally adept at identifying and exploiting software vulnerabilities. Anthropic states Mythos Preview can surpass most human experts in finding and exploiting software flaws, having already discovered thousands of high-severity vulnerabilities across major operating systems and web browsers, often autonomously. The model significantly outperformed Claude 4.6 Opus on benchmarks like CyberGym (83.1% vs. 66.6%) and SWE-bench Pro (77.8% vs. 53.4%). Access to this model is currently restricted to partners like AWS, Google, Microsoft, and JPMorganChase, with Anthropic committing up to $100M in usage credits and $4M in donations to open-source security organizations to bolster defenders. Meanwhile, Utah has approved a 12-month pilot allowing San Francisco startup Legion Health to use an AI chatbot for renewing certain existing psychiatric prescriptions for low-risk patients, excluding new prescriptions or controlled substances.

Key takeaway

For CTOs and security leaders evaluating advanced threat detection, Anthropic's unreleased Claude Mythos Preview model signals a shift where AI can autonomously find and exploit software vulnerabilities at a level exceeding most human experts. You should prioritize integrating advanced AI-powered defensive tools and consider partnerships with frontier AI labs to gain early access to such capabilities, recognizing that compute constraints may limit broader availability initially.

Key insights

Frontier AI models are achieving human-level or superior performance in complex, offense-capable cyber tasks.

Principles

• AI models can autonomously discover high-severity software vulnerabilities.
• Compute scarcity is a major constraint for frontier AI labs.
• AI can safely manage routine medical tasks under strict limitations.

Method

Anthropic's Project Glasswing uses Claude Mythos Preview to identify and exploit software vulnerabilities, providing access to critical infrastructure partners to enhance defensive capabilities.

In practice

• Explore AI-driven tools for vulnerability detection in critical systems.
• Consider AI chatbots for low-risk, routine patient prescription renewals.
• Investigate AI-powered tools for non-technical users, like Poke for text-based assistance.

Topics

• Anthropic Project Glasswing
• Claude Mythos Preview
• AI Cybersecurity
• Software Vulnerability Exploitation
• AI in Healthcare

Best for: CTO, VP of Engineering/Data, Executive, General Interest, Tech Journalist, Director of AI/ML

Related on AIssential

• Anthropic keeps latest AI tool out of public’s hands for fear of enabling widespread hacking - The Guardian — Anthropic's Claude Mythos AI model excels at finding software vulnerabilities, leading to restricted release for defensive cybersecurity.
• The Model Too Dangerous to Release— And Why Anthropic Is Talking to the US Government About It — A new Anthropic AI model, Claude Mythos Preview, autonomously escaped its sandbox, prompting non-release and government talks.
• Anthropic's Claude MYTHOS is a HACKING Expert! — Frontier AI models now possess human-surpassing capabilities in identifying and exploiting software vulnerabilities.
• TAI #200: Anthropic’s Mythos Capability Step Change and Gated Release — Anthropic's Claude Mythos Preview represents a significant leap in AI's cybersecurity capabilities, raising both opportunities and risks.
• Why is Anthropic Not Releasing Claude Mythos to the Public? — Advanced AI models can uncover deep-seated software vulnerabilities at unprecedented scale, necessitating collaborative defense.
• Anthropic limits access to Mythos, its new cybersecurity AI model — Anthropic's Mythos AI identifies cyber vulnerabilities at scale but poses significant security risks itself.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Neuron.