What CrowdStrike published
CrowdStrike's May through August 2026 adversary research documented four distinct threat actor campaigns where operators use agentic AI frameworks to automate portions of the attack lifecycle. The campaigns range from nation-state espionage groups using AI-assisted reconnaissance pipelines to financially motivated actors automating credential stuffing and initial access workflows with scaffolding derived from PentestGPT-style tooling.
What agentic AI changes in an attack
Traditional manual attacks require an operator behind a keyboard for each step: reconnaissance, vulnerability identification, exploitation, and lateral movement. Agentic attacks delegate these steps to an AI system that can iterate faster, run in parallel across targets, and execute at hours when human operators are unavailable. CrowdStrike's reporting notes that observed campaigns reduced operator time-on-keyboard for initial access phases from hours to minutes in several documented intrusions.
Campaign patterns worth knowing
Across the four documented campaigns, recurring patterns include automated OSINT pipelines that aggregate public-facing data about target employees and infrastructure before the first direct probe, credential spray workflows that adapt spray patterns in real time based on error responses, and post-compromise scripting where AI-generated payloads bypass signature-based detections by generating novel variants per target.
What defenders should consider
- Detection logic tuned for human-paced attacks may underperform against agentic attack cadences. Review detection rules that assume deliberate timing between steps.
- Credential spray defenses, including lockout policies and anomaly detection, need to account for adaptive spray patterns that vary timing and source to avoid triggering static thresholds.
- Endpoint and network telemetry is more valuable than ever as AI-assisted attacks increase scale. Ensure coverage is comprehensive rather than relying on sampled event streams.
The durable lesson
Agentic AI in the hands of threat actors is not a future scenario. Four documented campaigns in a single research window suggests this is an active operational shift, not a proof of concept. Security controls designed around human-paced threat actors will need calibration as agentic attack patterns become more common across the threat landscape.
Gigia Tsiklauri is a Security Architect and founder of Infosec.ge. Get in touch if you want to discuss agentic AI security or threat intelligence for your organization.