- Published on
New serious vulnerabilities spiked around release of Claude Mythos Preview
- Authors
- Name
- DeepHorus
The number of high-severity cybersecurity vulnerabilities has surged dramatically following recent advancements in artificial intelligence, with a record spike coinciding with Anthropic's announcement of its latest AI model's autonomous hacking capabilities. In April 2026, Anthropic revealed that its Claude Mythos Preview model could autonomously discover and exploit software vulnerabilities. This breakthrough prompted both Anthropic and OpenAI to initiate projects using frontier models to proactively secure critical software before malicious actors could leverage similar technology. The announcements marked a turning point in the use of AI for cybersecurity, raising both hopes and concerns about the speed of vulnerability discovery.
According to data from the Common Vulnerabilities and Exposures (CVE) system, the monthly count of high- and critical-severity vulnerabilities jumped more than 3.5 times in June 2026 compared to the previous record high before the Mythos Preview announcement. The spike represents an unprecedented acceleration in vulnerability reporting, far exceeding typical seasonal variations. This surge is directly attributed to the release of Claude Mythos Preview, as researchers and automated systems began exploiting its capabilities. The data underscores the model's effectiveness in identifying previously unknown flaws.
The sudden increase suggests that AI-powered vulnerability discovery is reshaping the cybersecurity landscape. While the modeling may help defenders patch flaws faster, it also raises concerns about the potential for misuse. The same technology could be employed by adversaries to discover zero-day exploits at scale, potentially overwhelming existing defense mechanisms. This dual-use nature of AI in cybersecurity demands careful management.
The industry is divided in its response. Some experts see this as a necessary step toward automated security auditing, improving the overall resilience of software ecosystems. Others warn that the rapid disclosure of vulnerabilities could overwhelm patching teams, leaving systems exposed for longer periods. Anthropic and OpenAI have emphasized their commitment to responsible disclosure and hardening critical infrastructure, but the pace of discovery continues to outstrip response capabilities.
As AI models continue to improve, the pace of vulnerability discovery is likely to accelerate further. The challenge will be to balance proactive defense with the risk of exposing unpatched systems to malicious actors. This trend underscores the urgent need for automated patch management, coordinated disclosure processes, and robust AI governance frameworks. The cybersecurity community must adapt quickly to prevent the same tools that secure our networks from being turned against them.
