Civil-society participation inside AI technical-standards bodies

A civil-society organisation gets staff seats inside the technical-standards bodies that write the documents regulators later adopt — the NIST AI Risk Management Framework, ISO/IEC JTC 1/SC 42, IEEE working groups, the IRTF Human Rights Protocol Considerations research group, ENISA technical guidance. The work is grinding: weekly drafting calls, edit-suggestions on hundred-page documents, plenary meetings, comment cycles. The output is wording inserted into a standard — that "fairness assessment shall include external review," that risk evaluations must "consider impacts on affected populations" — which then becomes load-bearing once any regulator references the standard.

An actor chooses this strategy because standards documents are quietly the most powerful regulatory artifact in the AI policy stack: a single technical line written by a working group of fifteen people ends up cited by a dozen agencies and embedded in millions of contracts. Industry has dominated these rooms by default because participation is unpaid, slow, and requires deep technical literacy; sustained civil-society presence is the only way the eventual standard reflects anything beyond a vendor consensus. The leverage-to-attention ratio is unusually high — a year of patient drafting can rewrite the default that a generation of compliance teams will follow.

It trades that ratio for capture risk and very high opportunity cost in scarce staff time. The rooms are designed to produce consensus, which means a civil-society delegate is structurally pushed toward compromise on positions the broader movement might hold firm on outside the room; success is also genuinely invisible — a sentence in an annex — which makes the work hard to fund, hard to celebrate, and easy to abandon when shorter-cycle campaigns demand the same people.

Verdict — high-leverage doctrine, severely under-staffed in the corpus

Civil-society participation inside technical-standards bodies is, in the corpus this team is mapping, a strategy with very high theoretical leverage and severely thin organisational adoption — concentrated, in practice, in two named individuals carrying organisational mandates rather than in staffed organisational programmes. The mapped adopters are Coding Rights (via Joana Varon's co-initiation of the IRTF HRPC research group and the RFC 8280 methodology that followed), ARTICLE 19 (via Amelia Andersdotter's sustained consultancy on human rights in IEEE 802, IETF, 3GPP, and ISO PC 317 IoT), and EDRi (via the same Andersdotter response-team role and the October 2018 EDRi-byline analysis of 5G standards). The corpus's clearest single Voice for the strategy — Amelia Andersdotter — runs the strategy's whole public thesis in two registers simultaneously: the Internet Policy Review peer-reviewed argument and the EDRi-and-engineering-venue civil-society translation.

The strategy is good at producing infrastructurally durable language inside high-leverage technical documents, weak at producing legible movement-side outputs, and structurally cut off from the public-mobilisation register the rest of the movement runs on.

Strong on its primary claim — that a sustained civil-society presence inside a standards body can shape the published output of that body (RFC 8280 as the worked case), and that the resulting document is then propagated downstream into civil-society methodology (the Oracle for Transfeminist Technologies carries IETF Human Rights Considerations methodologies as one of three named inputs) and into adjacent regulatory frameworks. The strategy's leverage is real and the published outputs are durable; the few corpus-mapped adopters are operating on a high-yield-per-staffer basis.

Weak on its scale-of-adoption claim. The corpus shows two named individuals carrying three organisational mandates between them. The standards bodies the strategy operates in — IEEE 802, IETF, IRTF, 3GPP, ISO/IEC JTC 1/SC 42, ITU-T, ENISA — together comprise hundreds of working groups and study items running on multi-year cycles, with industry-delegate counts running orders of magnitude above civil-society presence. The strategy's mapped civil-society capacity, taken across the entire corpus, sits at roughly two part-time consultant briefs against an industry presence numbered in the thousands. The leverage-to-attention ratio the strategy promises is real but unrealised at scale; the work is happening, but at a scale that cannot meaningfully contest more than a few specific draft clauses per year.

Structurally narrow in three ways. First, the adoption-shape gap: the corpus's adopters are individuals (Varon, Andersdotter) carrying organisational mandates, not organisations running multi-person standards-engagement programmes. This is structurally different from the climate movement's IRENA / IEA engagement, where named coalitions (the Climate Action Network, the IISD, the IEEFA) staff multi-person delegations across standards-body and intergovernmental processes; or from the labour movement's ILO standards engagement, where union confederations send full delegations year-round. The strategy as practiced in the corpus is on the single-consultant-with-mandate model, which is fragile (one person leaving collapses the engagement) and intrinsically scale-limited.

Second, the regulator-confusion gap: many corpus organisations that appear to be adopting the strategy are in fact engaging regulators (the UK ICO generative-AI guidance via Privacy International; the EU AI Act trilogue via EDRi-led coalitions; the US NIST AI RMF comment process) rather than standards bodies. The two are adjacent — regulators frequently cite standards-body outputs in their guidance, and the EU AI Act will reference harmonised standards — but they are structurally distinct: regulators issue binding rules and accept civil-society advocacy through formal consultation channels; standards bodies produce technical specifications and require sustained delegate presence through working-group drafting cycles. The corpus has many adopters of the regulator-engagement strategy (coalition-lobbying-of-binding-regional-regulation is heavily populated) and few of the standards-body strategy proper.

Third, the multilateral-policy-vs-technical-standards gap: organisations like APC that engage at the ITU and UNESCO on AI multilateral processes carry a policy-positions mandate that is adjacent to but distinct from the technical-standards drafting work the ITU-T sector also produces. APC's policy explainer names ITU among its engagement venues, but the engagement is registered as multilateral-policy advocacy rather than technical-document drafting; the lift required to convert that presence into sustained ITU-T study-group participation is non-trivial. The corpus's multilateral-policy adopters are not, on the record this team tracks, also operating as standards-body adopters in the strict sense the strategy names.

The strategy's verdict in the corpus is: the doctrine is right, the practitioners are few, and the gap between the strategy's theoretical leverage and its realised civil-society capacity is one of the largest the corpus surfaces in the entire ecology of make-AI-good tactics. The strategy is severely under-funded relative to its potential, structurally vulnerable to single-personnel-loss collapse, and disconnected from the public-mobilisation register on which the rest of the movement runs. The strongest analytical move for the movement looking at this verdict is not to abandon the strategy (the leverage is real and the worked instances are durable), but to staff it deliberately — to fund standards-body engagement at organisational scale rather than as a consultancy line-item, and to build the translation surface (counter-narrative, empirical-audit, open-letter) alongside it so that the standards-room work becomes legible to the broader movement and politically defensible to its funders.

Ecology

The strategy is the upstream tributary of coalition lobbying of binding regional regulation. The codified technical language regulators adopt in binding instruments — the EU AI Act's references to harmonised European standards, the NIST AI RMF's downstream propagation into US federal procurement, the Council of Europe Framework Convention on AI's referencing of technical guidance — is, with regularity, language that civil society either wrote in the standards room or failed to insert there. Standards-body work is what makes regulator-track work concrete; regulator-track work is what makes standards-body work binding. The two strategies sit upstream / downstream of each other and the corpus's clearest single argument for this relationship is the Andersdotter / Olejnik Internet Policy Review thesis, which is structurally an argument that the coalition lobbying of binding regional regulation strategy cannot succeed without civil-society participation inside technical-standards bodies as its upstream feed.

The strategy uses counter-narrative framing as its translation surface. The standards-room outputs are themselves illegible to the broader movement; the strategy's downstream effect on the movement depends on a parallel layer of civil-society publication that translates the standards-room observations into the broader register the rest of the movement reads. Andersdotter's Internet Policy Review article is the worked exemplar: the standards-room observation that 5G drafts embed vertical integration is reframed as a policy-and-regulatory argument that the EU's 1985 New Approach framework is structurally ill-suited to ICT. The counter-narrative work is what makes the standards-room work politically defensible to funders and movement peers, and the strategy is structurally dependent on having a counter-narrative-framing arm running alongside it.

The strategy sits in tension with parallel community research institution. The two strategies are alternative responses to the same problem — that AI technical expertise inside civil society is scarce and contested — and they take opposite institutional positions on it. Standards-body participation is institutional engagement (sit inside the room, shape the draft, accept compromise); parallel-research is institutional withdrawal (build the counter-expertise outside, refuse compromise, publish independently). Coding Rights is the corpus's clearest single case of an organisation running both at once: the Oracle for Transfeminist Technologies is parallel-research; the IRTF HRPC participation is standards-body engagement. The two strategies can be operated together — Coding Rights does — but they make different demands on staff and produce different kinds of legitimacy claims, and the corpus contains adopters of one but not the other (parallel-research adopters like Algorithmic Justice League and DAIR Institute do not, on the record this corpus tracks, run sustained standards-body engagement programmes).

The strategy reinforces and is reinforced by empirical audit and expose. Both strategies require deep technical literacy in civil society; both produce outputs whose authority derives from the rigour of the technical claim rather than the size of the constituency behind it. They share a personnel pipeline (the same scarce technically-literate civil-society staffers are eligible for either), and they share a downstream pathway into regulator-facing work (audit findings cited in standards-room arguments; standards-room language cited in audit reports). The two are structurally complementary — the audit produces the factual case that civil-society then carries into the standards room — but they also compete for the same scarce talent, which is one of the corpus's clearer instances of strategies that should coordinate at organisational scale but, on the record this team tracks, mostly don't.

The strategy is a tactical complement to open-letter collective signatory action, and could be more deliberately so. An open letter signed by hundreds of civil-society organisations carries political weight that no single delegate sitting in a working group can match; a sentence inserted into an IEEE draft carries technical durability that no open letter can produce. The strategies are structurally complementary, but the corpus shows few worked instances of orgs deploying them together — open letters tend to be addressed to regulators or platform CEOs, not to standards-body chairs, and the strategy's standards-room delegates tend not to be the same staff who organise open letters. The unrealised tactical combination — open-letter mobilisation timed to a standards-body comment-cycle deadline, with the letter's signatories backing the position the civil-society delegate is taking inside the room — is one of the more concrete unbuilt tactical surfaces the corpus's ecology surfaces.

The strongest competing strategy against this one is the AI industry's dominance of standards-body participation by default: hundreds-to-thousands of vendor engineer-delegates funded by the companies whose products the standards then specify, against a civil-society presence numbered in the dozens worldwide. The competition is not adversarial in the strategic sense (industry does not run a counter-strategy against civil-society standards-body engagement — it simply dominates the rooms by attendance); it is a structural ratio that the strategy must contest by simple multiplication of civil-society delegate capacity. The strategy's strongest unhedged claim — that civil-society participation inside technical-standards bodies can shape AI infrastructure at scale — is, on the corpus's verdict, structurally true but operationally severely under-resourced, and the most useful future move for the movement would be funding deliberate scale-up of standards-body engagement at organisational rather than consultant scale, with explicit recognition that the strategy is upstream of every other regulator-facing tactic the movement runs.