Wiki · Accountability completeness
Does any of it work? — the accountability ledger
For each of the 23 harms the catalog tracks, this maps which of four accountability lever categories act on it — binding law, soft/voluntary, technical standard, enforcement/litigation — which are absent, and whether any rigorous evidence shows the governance works. The pattern is stark: 23/23 harms carry a binding rule, yet 13 have no catalogued enforcement (“named but toothless”) and 23/23 have no rigorous evidence the governance works. Mean lever categories present: 2.8/4.
Honest scope
Which lever categories reach the harms
| Lever category | Harms reached | Share |
|---|---|---|
| Binding law | 23 / 23 | 100% |
| Soft / voluntary | 22 / 23 | 96% |
| Technical standard | 9 / 23 | 39% |
| Enforcement / litigation | 10 / 23 | 43% |
Per harm — least accountable first
| Harm (topic) | Levers | B | S | Std | Enf | Efficacy | Flags |
|---|---|---|---|---|---|---|---|
| Sovereign AI Doctrine | 1/4 | 5 | 0 | 0 | 0 | absent | named but toothlessno efficacy evidence |
| AI-Driven Worker Displacement | 2/4 | 3 | 6 | 0 | 0 | thin | named but toothlessno efficacy evidence |
| Compute + Model-Weight Export Controls | 2/4 | 1 | 3 | 0 | 0 | absent | named but toothlessno efficacy evidence |
| Compute-Threshold Reporting | 2/4 | 7 | 9 | 0 | 0 | absent | named but toothlessno efficacy evidence |
| AI in Education | 2/4 | 3 | 2 | 0 | 0 | thin | named but toothlessno efficacy evidence |
| International Coordination | 2/4 | 3 | 13 | 0 | 0 | absent | named but toothlessno efficacy evidence |
| National Security Carveouts in AI Regulation | 2/4 | 8 | 4 | 0 | 0 | absent | named but toothlessno efficacy evidence |
| Open-Weight Frontier Release | 2/4 | 5 | 7 | 0 | 0 | absent | named but toothlessno efficacy evidence |
| Technological Sovereignty | 2/4 | 5 | 6 | 0 | 0 | absent | named but toothlessno efficacy evidence |
| Agentic AI Governance | 3/4 | 9 | 7 | 2 | 0 | absent | named but toothlessno efficacy evidence |
| Catastrophic & Existential Risk | 3/4 | 7 | 13 | 2 | 0 | absent | named but toothlessno efficacy evidence |
| Environmental Impact of AI Training | 3/4 | 4 | 7 | 1 | 0 | absent | named but toothlessno efficacy evidence |
| Synthetic Content Provenance | 3/4 | 7 | 8 | 2 | 0 | thin | named but toothlessno efficacy evidence |
| Biometric Identification | 3/4 | 6 | 2 | 0 | 3 | thin | no efficacy evidence |
| AI in Criminal Justice | 3/4 | 4 | 2 | 0 | 1 | thin | no efficacy evidence |
| Development-Rights Framings | 3/4 | 5 | 6 | 0 | 1 | absent | no efficacy evidence |
| AI in Employment | 3/4 | 5 | 3 | 0 | 2 | absent | no efficacy evidence |
| AI in Healthcare | 3/4 | 4 | 3 | 0 | 1 | absent | no efficacy evidence |
| Deepfakes / Synthetic Content | 4/4 | 8 | 4 | 2 | 3 | thin | no efficacy evidence |
| Foundation Models / GPAI | 4/4 | 13 | 19 | 2 | 5 | absent | no efficacy evidence |
| Individual Redress | 4/4 | 15 | 10 | 2 | 11 | absent | no efficacy evidence |
| Training-Data Rights | 4/4 | 10 | 8 | 2 | 5 | absent | no efficacy evidence |
| Transparency Obligations | 4/4 | 18 | 21 | 2 | 9 | absent | no efficacy evidence |
How to read this
This is the “does any of it work?” sibling to /wiki/coverage-equity (which measures who is covered). A harm flagged named but toothless has a binding rule on the books but no enforcement case in the catalog; no efficacy evidence means no rigorous evaluation shows the governance reduces the harm (the per-topic evidence map carries the underlying findings). Both are prompts to scrutinise, not proofs of failure — see the honest-scope note above. Lever categories come from each instrument’s legal kind plus catalogued enforcement cases. Machine-readable at /wiki/accountability-completeness.json.