Unit 42 says it is already seeing early in-the-wild abuse of agentic and frontier AI, and one active case used an agentic framework to hit 50 applications and other enterprise weaknesses in under 10 hours. Palo Alto Networks’ threat intelligence team says that is a generational shift, because the work of finding and testing targets is now moving at machine speed.
The model is simple: instead of an operator probing one path at a time, AI can keep searching, chaining, and retrying across many applications on its own. That compresses reconnaissance and exploitation into the same short window, so an attack can widen before human triage or correlation catches up.
For organizations with broad application estates or supply-chain dependencies, the exposure is no longer just what a single attacker can do in a sitting. If defenses still assume reconnaissance is slow and noisy, they will undercount how far one campaign can spread before anyone notices.