Researchers reportedly used Claude Opus 5 to breach OpenAI's internal systems
An AI news roundup claims the security researchers carried out the breach within 72 hours.
TLDR
A September 18, 2026, AI news roundup claims security researchers used Claude Opus 5 to breach OpenAI's internal systems within 72 hours. The same roundup says Gary Marcus warned that large-scale hacking by AI agents deserves more near-term concern than runaway superintelligence.
Combined views
—
2 Sources, first seen ago
