Emerging story path

AI governance becomes measurable infrastructure

Oversight is being restated as numbers a third party could in principle check: monitored action coverage, escalation rates, compute allocated to safety, who may use which capability, and who gets to look inside.

Stage 1Emerging now3 entries
Stage 2Continuing story3 entries
Stage 3Established research path10 entries

Evidence status: Reported first-party facts plus labelled editorial synthesis. The reviewed record contains no independent replication of the reported rates.
Why it matters: Editorial synthesis: if oversight becomes a measured operating property rather than a policy statement, buyers and auditors can compare vendors on coverage, latency, escalation and independence rather than on assurances.
What would strengthen it: The reading strengthens if separate laboratories publish comparable oversight time series, and if outside evaluators reproduce the reported coverage and escalation measures.
What would disconfirm it: The reading weakens if independent auditors cannot reproduce the measures, if cross-lab definitions stay too inconsistent to compare, or if verified access does not improve legitimate task completion while raising misuse and data-governance failures.

Emerging now

Exact reviewed signals from the 2026-09-18 baseline: frontier-lab R&D, oversight and compute metrics; tiered life-sciences access; embedded evaluation.

Continuing story

Earlier reviewed signals on the same question.

Established research path

Historical milestones in the approved corpus.

Suggested alt text for the progression diagram: Three-stage diagram: established research on control, access control and evaluation leads to a continuing story of agent and safety evaluation signals, which leads to the September 2026 reviewed signals on measured AI oversight.