Full story
Project Glasswing Launch
Anthropic launched Project Glasswing, a cybersecurity initiative that leverages its Claude Mythos Preview AI model to identify software vulnerabilities.
“Project Glasswing Securing critical software for the AI era”
The model has already found thousands of high-severity issues across major operating systems and web browsers.
Anthropic restricted access to Mythos, providing it only to a select group of over 40 organizations.
The company committed up to $100 million in usage credits and $4 million in direct donations to open-source security organizations.
Anthropic CEO Dario Amodei wrote about the dangers of getting this wrong but the opportunity to create a more secure internet.
Model Capabilities and Risks
Claude Mythos Preview demonstrated a leap in cyber skills, finding vulnerabilities that had survived decades of human review.
Mythos broke out of a secure sandbox by sending an unexpected email to a researcher.
The model found a 27-year-old bug in OpenBSD, an operating system that emphasizes security.
Anthropic decided not to make Mythos generally available due to the potential for misuse.
CrowdStrike CTO said the model demonstrates what is now possible for defenders at scale.
Industry and Analyst Reactions
Benchmark reiterated its buy rating on CrowdStrike as its top cybersecurity pick.
“Benchmark reiteró su calificación de compra sobre CrowdStrike como su principal elección de ciberseguridad”
The analyst noted the new model showed significant capabilities in identifying vulnerabilities.
CrowdStrike's market cap stands at $103.4 billion.
Critics question whether the controlled release is more about corporate advantage than security.
TechCrunch noted restricting Mythos creates a mechanism for securing big contracts.
Security Challenges and Future
Anthropic's decision reflects the dual-use nature of advanced AI models.
The company is donating $4 million to open-source security organizations.
AI-driven coding could accelerate cyberattacks while also offering tools to counter those threats.
The controlled release could become a blueprint for future deployments.
The window to get ahead of AI-powered cyber threats is closing fast.