This is a scoped research dossier, not a completed systematic review or an independently tested result. It identifies methods, questions and source trails for future reporting.
Build a reliable task set
Report datasets, versions, scoring rules, contamination risks and representative negative cases. Hidden rubrics should not hide measurement design.
Measure interventions
Useful metrics include success, time, expense, policy violations, false completions and how often a human had to rescue the system.
Adversarial scenarios
Include tool failures, prompt injection attempts, stale context and cross-agent confusion. Separate detection accuracy from ability to recover.
What would count as evidence?
Create a versioned evaluation card with tests, failures, budgets and confidence limits.
Documents to examine
- NIST AI RMF
- OWASP — Excessive Agency
These are starting points, not claims that every document has been independently reproduced.
Read our cited field note →Edition 1.0 · 09 October 2026
Initial research brief published. No earlier revisions or submitted public corrections are claimed.
Suggest a documented correction ↗