Athena AI to Begin Disclosing 50 Vulnerabilities This Month, Joins Anthropic's Project Glasswing

Athena's Initial Vulnerability Disclosures

Athena, an AI system focused on vulnerability discovery, is set to begin disclosing its first 50 findings starting September 28. This marks the initial public release of vulnerabilities identified by the system, which has been operational since June.

Since its inception, Athena has identified more than 40,000 vulnerabilities across over 500 software projects. Of these findings, 42% are classified as critical or high severity, and the system has already produced over 2,000 patches. The primary challenge has been the responsible disclosure of these vulnerabilities at an unprecedented scale, rather than the generation of fixes.

Collaboration with Project Glasswing

Athena has joined Anthropic's Project Glasswing, a collaborative effort aimed at enhancing vulnerability discovery. This partnership will allow Athena to contribute its capabilities and further refine its methods for identifying security flaws in software systems.

Open-Sourcing Infrastructure for Disclosure

To support its large-scale vulnerability disclosure and patching operations, Athena has developed specialized infrastructure. This infrastructure will be open-sourced, providing other organizations and researchers with tools to manage the intake, processing, and responsible disclosure of security findings generated by AI or other automated methods.

Addressing the Consumption End of AI Risk

The initiative focuses on the 'consumption end' of AI risk, which involves how vulnerabilities discovered by AI models are handled and mitigated. This includes improving disclosure processes, patch pipelines, and the software itself to absorb the output of AI models. The goal is to defend against potential damage from AI-discovered flaws in critical infrastructure like power grids and hospital networks, which often run on older, exploitable software.