Microsoft Tests Mythos to Identify and Mitigate Vulnerabilities
Microsoft Tests Mythos to Identify and Mitigate Vulnerabilities
Publish Date: 2026-04-22 16:36:00
Source Domain: www.pymnts.com
- Microsoft partners with Anthropic and others in Project Glasswing to test the Claude Mythos Preview model, aiming to identify and mitigate cybersecurity vulnerabilities.
- Evaluations show that Mythos has significantly improved over previous models, as tested under Microsoft’s open-source benchmark, CTI-REALM.
- To keep pace with AI-driven threats, Microsoft plans to integrate advanced AI, such as Claude Mythos Preview, into its Security Development Lifecycle (SDL).
- Microsoft emphasizes a collaborative, proactive cybersecurity stance, focusing on patching, reducing exposure, and leveraging AI for enhanced security.
- Anthropic revealed an initiative called Project Glasswing allowing select partners access to Claude Mythos Preview for defensive cybersecurity; partners include Amazon, Apple, and cybersecurity firms like Crowdstrike and Palo Alto Networks.
- Anthropic is investigating reports of unauthorized access to Claude Mythos Preview through one of its third-party vendors.
