Palo Alto Networks' Unit 42 published a detailed account of a ransomware intrusion in which AI agents handled reconnaissance, credential theft, and lateral movement with no human operator directing individual steps. The attacker set objectives; specialized agents executed, shared results, and adapted in real time. What compressed the timeline also left a record: Unit 42 reconstructed the entire operation, including which of over 50 MITRE ATT&CK techniques were used at each stage, because the agentic workflow generated a legible sequence of actions rather than the improvised, inconsistent trail a human red team leaves behind. The attack itself is the headline most outlets ran with — a coordinated intrusion that would normally take human operators around two weeks, compressed into under ten hours. Specialist pivot agents confirmed reach into cloud, identity, container, SaaS, and build-pipeline environments, and the attacker subverted CI/CD workflows to steal cloud keys, all while orchestration traffic hid inside legitimate activity. The less-discussed detail is what made the postmortem possible at all. Because each agent's actions were logged as discrete, structured steps — mapped cleanly onto attack frameworks — investigators could produce a step-by-step workflow diagram of the entire intrusion. A separate July incident involving the Hermes Agent framework surfaced for a similar reason: the campaign was only discovered because the attacker's own tooling exposed session logs, API keys, and exploitation records on an open file server. This is a backhanded case for attestation infrastructure. Autonomous agents, whether defensive or offensive, produce far more granular execution records than human operators do, simply because they're built to log their reasoning and actions in structured form. Unit 42's advice to defenders is to match machine speed with automated response — isolating accounts, freezing pipelines, and revoking credentials simultaneously rather than sequentially. That kind of automated response depends on trustworthy, real-time attestation of what agents on your own network are doing, not just what attackers' agents left behind after the fact. The uncomfortable symmetry: if attacker agents generate audit-grade evidence of their own operations by default, defenders lacking equivalent instrumentation on their legitimate agent fleets are working from a worse evidentiary position than the people attacking them. Accountability infrastructure isn't just a compliance checkbox: it's the difference between reconstructing an incident in hours versus never fully understanding what happened.