Epoch AI recorded a massive jump in reported security flaws following the release of Anthropic‘s Claude Mythos Preview model in April. In June 2026, twenty-one organisations reported approximately 1,500 high-severity and critical vulnerabilities, which is more than 3.5 times the previous monthly record. This surge follows the announcement that the model can autonomously hunt for software bugs. Anthropic’s Glasswing programme has reportedly uncovered over 10,000 high-severity or critical issues to date, some of which remain unpublished. OpenAI’s Daybreak programme is likely contributing to the increase as well. The data suggests this spike reflects a wave of discoveries driven by artificial intelligence tools rather than human researchers alone.
The shift changes how security teams allocate resources and validates automated testing at scale. Companies must adjust their patching cycles because new threats appear faster than traditional teams can analyse them. The volume of findings also indicates that AI models may miss context or generate false positives alongside real bugs.
* Anthropic’s Glasswing programme has found over 10,000 critical issues
* June 2026 saw 21 organisations report 1,500 new CVEs
* OpenAI’s Daybreak programme is adding to the reported surge
