Anthropic Opens Claude Security Tools with Project Glasswing Update

What is Project Glasswing?

Anthropic's Project Glasswing is the company's initiative to use its Claude AI model for offensive security research. The Mythos Preview program demonstrated Claude's ability to find and exploit vulnerabilities in software systems — and now Anthropic is sharing some of those tools more broadly.

New tools now available to qualifying customers

Upon request, qualifying customers can now access skills, a Claude harness, and a threat model builder. These are the same tools Anthropic used internally during the Mythos Preview program to identify zero-day vulnerabilities in open-source software.

A dashboard for disclosed vulnerabilities

Anthropic has also published a public dashboard of open-source vulnerabilities discovered by Mythos Preview. This transparency move allows the broader security community to track what Claude found and verify the fixes.

Why this matters for enterprise security

AI-powered security tools represent a paradigm shift in cybersecurity. Instead of relying on human pentesters who can only cover so much ground, AI agents can systematically scan vast codebases for vulnerabilities at machine speed.

The ethical balancing act

Anthropic's work with the Pentagon has drawn scrutiny about the risks of autonomous warfare. The Glasswing update focuses on defensive security — finding and fixing vulnerabilities — which is a more universally accepted application of AI security capabilities.

Frequently Asked Questions

Q1: Who qualifies for these security tools? A1: Anthropic says "qualifying customers" can request access, though specific criteria haven't been publicly detailed. Expect vetting to prevent misuse.

Q2: How does Claude find vulnerabilities? A2: Claude analyzes source code, identifies patterns associated with security flaws, and can generate proof-of-concept exploits to confirm vulnerabilities exist.

Q3: Is this different from existing security scanners? A3: Yes. Traditional scanners use pattern matching and known vulnerability signatures. Claude can reason about code logic and find novel vulnerabilities that pattern-based tools miss.

Stay ahead of the AI curve. Follow @AiForSuccess for daily insights.