New serious vulnerabilities spiked around release of Claude Mythos Preview (epoch.ai)

In April 2026, Anthropic introduced its latest internal model, Claude Mythos Preview, which is designed for autonomous discovery and exploitation of cybersecurity vulnerabilities. This announcement sparked significant concern within the AI/ML community, as it was followed by a striking increase in reported Common Vulnerabilities and Exposures (CVEs). By June, the number of high- and critical-severity vulnerabilities surged more than 3.5 times compared to the prior monthly records, indicating a troubling trend in the exploitation of AI models. The implications of this spike are significant for cybersecurity protocols and the ongoing efforts to leverage advanced AI capabilities for protection rather than harm. Both Anthropic and OpenAI have since mobilized to fortify critical software systems before malicious users can exploit these advanced AI models for nefarious purposes. The events highlight a critical arms race in the realm of AI security, where the tools intended to safeguard technology could also be repurposed for detrimental objectives if proactive measures are not implemented promptly.