Standing AI accountability index

Movement organisations build and sustain a standing public index — a rolling scorecard ranking AI vendors, deployers, or governments on the movement's chosen accountability dimensions, updated on a defined cycle (annual, biennial), published with a comparative league table, and cited by journalists, procurement offices, and investors during the year that follows. The vehicle is a documented methodology (weighted dimensions, verification procedure, data sources, contestation window for scored entities), a permanent staff role owning the annual publication cycle, and a media strategy that lands the release with sustained press coverage. The reference model is Ranking Digital Rights' Corporate Accountability Index (governance, freedom of expression, privacy for the 26 largest platform/telecom companies); adjacent-movement analogs include Corporate Human Rights Benchmark, CDP climate disclosure ratings, RSF Press Freedom Index, and the Human Rights Watch World Report.

An actor chooses this strategy because it converts episodic movement pressure into standing infrastructure that compounds. A single audit lands once; an index accumulates comparative history over years, exposing not just where a vendor sits today but the trajectory it has moved along, and creates a reputational asset (a good ranking) or liability (a poor one) the vendor must defend to its investors, its board, and its next hire. The methodology itself becomes a movement artifact — a specific operational definition of what "accountable AI" means, published, reviewable, contestable — and its adoption by third-party consumers (journalists, investors, regulators) launders the movement's normative claims into institutional facts they cite when acting. The strategy also creates a defined interlocutor cycle with the scored vendors: the annual verification window becomes a scheduled negotiation in which the vendor must justify its practices to a movement-run authority.

It trades the sharpness of a single-target expose for the diffuseness of a comparative scan. An index that ranks twenty vendors on twelve dimensions cannot produce the tight, singular news story a targeted audit produces, and its outputs are frequently absorbed as "AI report cards" that generate one round of coverage before returning to the movement's own communications archive. The strategy is also methodologically brittle: methodology contestation from scored vendors is guaranteed and time-consuming; the choice of dimensions and their weights is itself a political act that shapes what "accountable" means; and an index that scores dimensions on which the movement's own coalition disagrees fractures the coalition around the release. Sustained credibility depends on visible independence from any single funder or campaign — an index that appears to serve one funder's interest loses its authority as a neutral comparative frame.

Ecology

Distinct from empirical audit and expose. Audit is episodic and adversarial: a single deployment, a single vendor, a single test that exposes a specific harm. An index is periodic and comparative: a defined roster of targets ranked on a stable set of dimensions across years. The two work together — the audit is the news story that becomes an index datapoint the next year — but their organisational forms and rhythms differ. An audit organisation is a small team of investigators; an index organisation is a methodology committee, a data team, and a permanent publication cycle.

Distinct from community-defined benchmarks and standards. Benchmarks measure AI system performance on specific tasks — how well a model classifies faces, how well a chatbot handles a harm scenario. Indexes measure corporate / institutional behaviour — how the vendor governs its systems, discloses its practices, treats its workers, remediates harm. The two are complementary layers of the "how do you know" question: benchmarks tell you how the AI performs, indexes tell you how the organisation deploying it behaves.

Distinct from public AI incident registry building. An incident registry catalogues events — individual harms as they occur — accumulating a searchable factual record. An accountability index derives scores — aggregate assessments — of the actors implicated. The registry is inputs; the index is judgments. A mature ecosystem runs both: the registry feeds the index with the year's events; the index consumes the registry into a comparative ranking.

Feeds counter-narrative framing. An index release is a scheduled communication moment the movement can plan messaging around: the framing organisations time their year's argument to the index publication so the ranking arrives inside a movement-set frame rather than being interpreted by the vendor's PR.