AI agents are making security’s breakout-time metric obsolete, and no one agrees on what replaces it
An AI agent operating inside an Argentine healthcare network changed approach just 29 seconds after its first attempt was stopped by firewalls or access controls, turning to the domain controller through SQL, according to CrowdStrike. Within 17 minutes, it had extracted five credential hashes from that controller. CrowdStrike said the attacker behind the operation compromised at least 15 healthcare organizations in Argentina and stole credentials and medical records from at least one of them. The company’s October 21 report highlights four previously undisclosed attackers tracked between May and August, all linked to Claude Code, and argues that agentic AI is making baseline offensive cyber activity more accessible. In each case, the offensive agent framework was exposed through open directories. In the Argentine case, a “Supervisor” agent selected the next action while an “Executor” agent carried it out. Anthropic was contacted and did not respond by press time.
CrowdStrike says human operators issue dozens of commands an hour, while AI agents can issue about 1,500, making the difference one of speed, scale, and orchestration rather than new exploits. The company also says it has already seen three times as many agentic intrusions in the second half of 2026 as in the first half. That acceleration is pushing the industry away from “breakout time,” once a key benchmark for how quickly an intruder can move from one machine to another. CrowdStrike’s own leadership now says that metric no longer reflects the reality of modern attacks, with internal debate ongoing over what should replace it. Other security teams are pointing to clocks that focus more on detection, response, revocation, and remediation, since attacks can unfold in minutes while patching and containment often take far longer.
