AI-assisted discovery made patch latency a design constraint.
What happened. Google Cloud and Mandiant argued that capable models were compressing the vulnerability-discovery and exploitation cycle, while the same techniques could help defenders review code and triage findings.
Impact and confidence
The direction is supported by disclosed research and vendor experience; the rate of real-world attacker adoption remains uneven and should not be generalized from demonstrations.
Defensive takeaway
Reduce exposed inventory, improve update verification, and require human validation for model-generated findings before they trigger consequential changes.