Frontier models change the attacker timeline. They can find vulnerabilities, chain exploits, and adapt to defenses fast enough that patching alone no longer closes the gap. The standard response misses the real shift: defenders are no longer racing disclosure, they are racing machine-speed discovery and weaponization.
Unit 42 says frontier AI, including Anthropic Mythos, can do the equivalent of a year of manual penetration testing in under three weeks. It also says attackers already begin scanning new CVEs in under 15 minutes, and that frontier AI will compress that window further. CSO’s guidance points to the operational answer: behavioral detection, automated triage, and faster containment, because alerts and manual review cannot keep pace.
The forward risk is not just more findings. It is that exploitable weaknesses will be discovered and used before normal patch cycles even start, which makes AI-assisted detection and response part of baseline defense rather than an advanced capability.