Cybersecurity disclosures, measured by Common Vulnerabilities and Exposures (CVEs), surged in June 2026, exceeding previous records by over 3.5 times. This dramatic increase follows Anthropic's April announcement that its Claude Mythos Preview model could autonomously find software flaws. The company's Project Glasswing partners, including Microsoft and Google, leveraged this capability to identify and patch vulnerabilities before the model's public release.

Since its inception, Project Glasswing claims to have uncovered over 10,000 high- or critical-severity vulnerabilities. This proactive approach mirrors efforts by OpenAI with its Daybreak product, aiming to harden critical software before malicious actors can exploit similar AI capabilities. The trend suggests a new arms race where AI is used for both offense and defense in cybersecurity.

The spike in reported CVEs in June, with an estimated 1,500 high- and critical-severity disclosures from notable organizations, directly correlates with the deployment of advanced AI vulnerability discovery tools. This period marks a significant shift in how software security is managed, driven by the capabilities of frontier AI models.