📌 When human oversight becomes a compliance requirement "We had humans in the loop" is a description of a process. "Prove it" is a demand for evidence, and evidence has properties good intentions do not. It has to name specific people, show they were distinct real humans rather than sybils or one-shot contractors, show their judgement held up over time, and make every contribution attributable, timestamped and tamper-evident. The ground is already moving. The EU AI Act requires human oversight for high-risk AI under Article 14, and the frontier labs themselves are warning that recursive self-improvement could quietly drop the human from the loop. The moment oversight has to satisfy an auditor, attestation ("trust us, qualified humans reviewed this") is not enough. You need provenance: a record someone who does not trust you can check. Run the self-check 👉 https://ont.io/news/human-oversight-documentation/