Amazon Joins Project Glasswing to Test Anthropic’s Mythos AI for Security Flaws
Amazon partners with Apple, Microsoft, Cisco and others on Project Glasswing to test Anthropic’s unreleased Mythos AI model for cybersecurity flaw detection. Mythos has already uncovered a 27-year-old bug in critical internet software and a 16-year-old vulnerability in video game code, with findings set to inform industry guardrails.
1. Project Glasswing Partnership
Amazon has joined a consortium including Apple, Microsoft, Cisco and other firms under an initiative called Project Glasswing to gain early access to Anthropic’s unreleased Mythos AI model. The goal is to hunt for vulnerabilities in member products and share findings across the industry to reinforce collective cyber defenses.
2. Mythos Model Capabilities and Findings
Anthropic’s Mythos is a general-purpose AI model repurposed for security testing, which has already identified a 27-year-old bug in foundational internet software and a 16-year-old vulnerability in widely used video game code—issues that evaded traditional automated scanners.
3. Implications for Amazon
By leveraging Mythos, Amazon aims to strengthen its cloud and retail platforms against emerging cyber threats, inform internal guardrails for advanced AI deployments and potentially integrate the model’s insights into AWS security offerings.