Agent Testimony
Discovery law already handles digital records aggressively in civil litigation. Federal Rule of Evidence 803(6) already provides a path for admitting regularly kept operational records. Courts already wrestle with AI-adjacent evidence reliability — the NCSC published a 2025 bench guide for judges handling AI-generated evidence.
Reasoning Chain
PDM: 45/60 (Imminent). Capability: 9/10 — Agents already emit rich action traces in production systems. Compute: 8/10 — Storage and logging are cheap; retention is mostly policy-bound. Research: 8/10 — Active legal analysis on AI evidence standards. Open Knowledge: 7/10 — Judges and litigators are learning, but unevenly across jurisdictions. Infrastructure: 7/10 — Existing e-discovery tools can ingest logs; chain-of-thought handling is unsettled. Governance: 6/10 — No agent-specific evidence doctrine yet; doctrine can still be built via case law.
Watching
Milestone A (Q4 2026–Q2 2027): first clearly documented subpoena/order targeting agent operational records in a mainstream US civil or regulatory case. Milestone B (2028–2029): published ruling that addresses admissibility boundaries for agent-generated materials (e.g., logs vs reasoning text vs derived summaries).