Frontier AI flaw-hunters warn attackers could gain broad access within months
Palo Alto Networks warns that frontier AI cyber models like Mythos and GPT-5.5 can rapidly identify and chain multiple flaws, finding vulnerabilities across hundreds of products and generating working exploits in internal tests (about 70% success) even as most flaws aren’t actively exploited in the wild. False positives run around 30%, and human expertise remains essential. The firm urges a four‑pronged defense—patch quickly, reduce internet exposure, deploy real‑time detection, and integrate AI/automation into security operations—as policy makers weigh limits on powerful AI in cybersecurity.
