Ofgem figures reported by the BBC this month put the electricity requested by data centre projects queuing for a grid connection at 125GW, up from 41GW a year earlier, and in Scotland campaigners at 23 separate proposed sites have stopped fighting individually and started drafting national rules. Neither of those facts appears in any AI strategy document as a risk. They belong to a category most organisations still file under public relations: people saying no. A new project called the AI Resist List, described in Tech Policy Press by Petra Molnar, argues that this category has been mismeasured for years, and it makes the case by simply writing the refusals down.
Strategic Insight: Refusal is usually treated as sentiment, which is a thing you manage with messaging. The list reframes it as behaviour, which is a thing that shows up in your consent timeline, your credit terms and your staffing. Sentiment is cheap to ignore. Behaviour is not.
What the project actually did
The list is a public, collaboratively assembled database of documented acts of opposition to the AI industry, built by researchers, journalists and scholars spread over seven time zones, with backing from the Distributed AI Research Institute, We and AI, and York University’s Refugee Law Lab. Verónica Martínez and Wael Qarssifi, both fellows of the Migration and Technology Monitor, carried out the initial scoping; both report on surveillance from inside communities that live under it.
The methodology is worth noting, because it is unusual. Every initiative named was approached before publication, descriptions were reviewed repeatedly for accuracy and for the safety of the people named, and nothing went in without consent, with a single exception where the individual could not be contacted but was already documented at length in litigation. Among entries with a stated region, close to six in ten sit in the Global Majority.
Entries are sorted under a four-part scheme its authors call the Countering AI Inevitability Framework: resisting, refusing, reclaiming, reimagining. The taxonomy does real work, because it stretches the definition of opposition well past the thing boards currently watch for. A lawsuit is legible. A union petition is legible. A Chilean municipality building a website that looks like a chatbot and routing the queries to actual residents, who fielded 25,000 prompts over twelve hours, sent from 68 countries, is not something anybody’s risk function was scanning for. Neither is a browser plugin that slows an assistant’s replies to a crawl so the user notices what they are waiting for.
That breadth is the point. Molnar’s argument, following Karen Hao’s reporting, is that inevitability is a claim rather than a fact, and that the claim survives partly because the counter-evidence has never been gathered in one place.
Why a global list is a British problem
The obvious reading of a project like this is that it concerns somewhere else: cobalt in the Democratic Republic of Congo, annotation work in Nairobi, a hyperscale campus near the US southern border. The United Kingdom is on the list too, and more to the point, the four demands the project distils from its entries are already live here in forms that have nothing to do with activism.
| What it measures | Figure | Why it lands on a board agenda |
|---|---|---|
| Data centre grid connection requests (Ofgem, year to Aug 2026) | 41GW to 125GW | Contention for a scarce connection is now decided partly by which schemes attract objection |
| Scottish sites with organised local opposition | 23 | Objectors are coordinating nationally, so a local consent strategy meets a national opponent |
| Scottish ministerial notification threshold | 50MW, flagged within 7 days | Political visibility of a scheme now begins before the planning committee does |
| UK public prioritising AI safety over speed | 89% | The permissive public your growth case assumes does not exist in the polling |
| UK adults who feel they have no meaningful say on AI | 60% | Exclusion, not technophobia, is what converts unease into organised opposition |
The last two come from the Ada Lovelace Institute’s Great Expectations briefing, which we analysed in our review of the UK trust gap. The Scottish figures come from our reporting on objectors organising for national rules and on the new ministerial notification direction; the Ofgem number is from our coverage of Welsh residents contesting Newport and Cardiff sites.
Critical Context: The financial system reached this conclusion before most boards did. Lenders have begun treating community opposition as an input to data centre credit risk rather than as reputational noise. When resistance is priced, it has stopped being a matter of opinion.
The four demands, read as costs
Molnar draws four cross-cutting demands out of the entries. Each one has a British price attached, and in three cases the invoice has already arrived.
Transparency. The recurring ask is elementary: which systems are being built, by whom, drawing on what, and at whose cost. In Japan, a workers’ union took IBM to a Tokyo labour commission to force disclosure of the AI data used in setting wages, arguing in its petition that “Humans place too much faith in decisions taken by AI”. The case settled in the union’s favour. The UK analogue is not exotic. It is the emissions figure in a planning submission that a campaign group recalculates, or the water abstraction estimate that turns out to assume a cooling design nobody has committed to. Disclosure is dismissed as a weak instrument right up until a contested number stalls a determination.
Participation. More than a thousand Amazon employees put their names to an open letter addressed to their leadership during 2025, carrying three demands: “no AI with dirty energy, no AI without employee voices, and no AI for violence, surveillance, or deportation”. Set that beside the Ada Lovelace finding that 60% of UK adults feel they have no meaningful input into government decisions on AI. Consultation that arrives after the decision has been taken does not reduce opposition; it supplies opponents with a documented grievance and a date.
The supply chain. The list traces AI from the mine to the annotation desk to the finished product, deliberately, precisely because industry framing has succeeded in making those links feel abstract. Friends of the Congo works with families pursuing legal accountability from technology companies over children’s deaths in cobalt mines. Nairobi’s Data Labelers Association, approaching 900 members, organises for contract transparency and for recognising the emotional labour that training these systems demands. For a UK buyer this is a procurement question rather than an ethical one: your supplier’s supplier is where the enforceable claim eventually lands.
Who controls the infrastructure. The campaigns against a hyperscale site in New Mexico and a Google facility in Uruguay are, on the surface, environmental disputes. Read together they are an argument about whether land, water and power for AI are public resources or private assets. Britain is having exactly this argument, in the grid queue and in Scottish planning, and it has not been resolved by anybody’s national strategy.
Strategic Reality: Three of these four demands are already enforceable in the UK through existing machinery. Transparency arrives via planning and environmental disclosure, participation via statutory consultation and union recognition, infrastructure control via consent and connection. None of them requires new AI legislation to bite.
What this does to the people inside your organisation
Labour is what recurs most insistently across the entries, and it does not stay outside the building. Mental health clinicians at Kaiser Permanente in California devoted 24 hours of a five-day hunger strike specifically to the prospect of AI entering mental healthcare, part of a work stoppage lasting six and a half months. Their objection was not to software. It was to clinical judgement being replaced by algorithmic triage, and to the compression of time per patient that follows.
British organisations should expect the same objection in a milder register, and should expect to mishandle it. We have written about the employee who refuses AI on principle and why UK employers tend to brace for the wrong legal test. The list adds a second observation: the objections that eventually become organised are the ones that were first raised individually and dismissed as attitude.
| Group | What they are actually contesting | What reduces the friction | Signal that it is working |
|---|---|---|---|
| Local communities near sites | Consent given without disclosure of water, power and emissions assumptions | Publishing the contested numbers before the campaign calculates them | Objections cite your figures rather than dispute their existence |
| Employees and unions | Judgement transferred to a system without their input | Involvement at design of the mandate, not at the point of refusal | Objections arrive through a named channel, not a grievance |
| Customers and clients | Service quality degraded by unannounced AI substitution | Disclosing where AI sits in the delivery chain | Complaints name the process rather than the outcome |
| Lenders and insurers | Opposition risk that was not underwritten | Evidence of consent strategy in the credit file | Opposition does not surface first at diligence |
Hidden Cost: An objection handled informally has not been resolved. It has been removed from your records while remaining in the objector’s, which means the first written account of your conduct is theirs.
What to do about it
💡 Implementation Framework: Treating refusal as a governed risk
Phase 1: Locate it (weeks 1 to 4)
- Identify every point where your AI programme depends on a permission somebody could withhold: consent, connection, contract, recognition, custom
- Ask which of those permissions currently has a named owner
- Log objections already raised and resolved informally
Phase 2: Cost it (quarter 1)
- Put a delay figure on each permission failing, not a probability
- Establish which contested numbers a competent opponent could recalculate
- Brief the board on opposition as a schedule risk rather than a communications one
Phase 3: Reduce it (year 1)
- Publish the assumptions before you are asked for them
- Move consultation ahead of the decision it concerns
- Extend supplier diligence to the annotation and extraction layers
For organisations without infrastructure exposure
- Audit where AI substitution is undisclosed: the customer objections that escalate are usually about finding out afterwards, not about the technology itself.
- Give principled refusal a route: a named, non-disciplinary channel converts an organising grievance back into a manageable conversation.
- Read your own polling honestly: internal surveys reporting near-universal enthusiasm are measuring what staff think is safe to say.
For organisations building or buying capacity
- Publish the contested numbers first: emissions, water, jobs. A figure you disclose is a figure you control the framing of.
- Assume national opposition to a local scheme: the Scottish campaigns show coordination arriving faster than consent processes assume.
- Underwrite opposition explicitly: if your lender is pricing it, your own model should already have.
For organisations already in a contested process
- Separate the answerable objections from the positional ones: they need different responses, and conflating them loses both.
- Stop treating settlement as silence: an agreement that resolves one site does not resolve a campaign now organised across many.
- Document your own reasoning: the proportionality of your decision is judged on the record you made at the time, not the one you assemble later.
Resource Reality: Mapping the permissions your programme depends on is roughly a fortnight of one person’s time, plus a standing quarterly review. It is cheaper than a single quarter of delay on a contested determination.
Four things this makes harder than it looks
The comfort of the reasonable objection
Organisations answer the objections they can answer. A dispute over an emissions figure is tractable, so it gets attention, whilst an objection to the premise gets categorised as unreasonable and left alone. The unanswerable objection is the one that recruits.
Mitigation: Track objections by whether they are growing, not by whether you can rebut them. A position you cannot argue away is a position you need to negotiate with.
Consultation that documents its own emptiness
Statutory consultation creates a record. If the decision was taken beforehand, that record now shows it, and it will be read back to you in an inquiry or a tribunal. The exercise designed to reduce risk becomes evidence.
Mitigation: Consult on a question that is genuinely open, or do not consult. Retrospective consultation is worse than none, because it is written down.
Supply chain distance is not legal distance
The gap between a UK buyer and a cobalt mine or an annotation contract feels like insulation. It functions as insulation only until a claim is brought, at which point the relevant question is what you knew and what you asked.
Mitigation: Ask the questions in writing, at procurement. The cost of asking is small; the cost of never having asked is that you cannot show you did.
Adoption metrics conceal refusal
A programme reporting high adoption and no objections is usually not a programme without objections. Refusal that has nowhere to go does not disappear, it relocates, generally to somewhere you find out about later and publicly.
Mitigation: Report challenges raised and objections logged alongside adoption. Both being zero is a finding, not a success.
Reality Check: None of this stops a determined campaign, and it is not meant to. It changes what a campaign is arguing about, from your conduct to the merits of your scheme. That is a materially better argument to be having.
The part worth keeping
The project makes no claim that resistance wins. It claims that resistance exists, at scale, on every continent, in forms ranging from litigation to an artist’s browser plugin, and that its invisibility in mainstream coverage has been doing quiet work for the industry. For a UK organisation the strategic content is narrower and more useful than the politics. Opposition is no longer an externality. It has entered the grid queue, the planning direction, the credit committee and the union room, and it arrives on a timetable your programme plan does not currently model.
Three things carry the weight:
- Refusal is a schedule risk: it consumes time before it consumes reputation, and time is the thing your business case is least able to absorb.
- The demands are procedural, not ideological: transparency, participation, supply chain visibility, control of infrastructure. Every one of them is answerable with process rather than persuasion.
- The record you make now is the record that gets read: in an inquiry, a tribunal or a diligence process, the contemporaneous document decides it.
A better question than “how do we bring people with us”
That framing assumes the destination is settled and only the journey is in dispute, which is the inevitability claim restated as stakeholder engagement. It also predicts the wrong opposition, because it treats disagreement as a deficit of understanding.
The more useful question is which permissions your AI programme actually depends on, and what happens to the timeline if any one of them is withheld. That question has an answer, the answer is a number, and the number belongs in the business case rather than in an appendix about community relations.
Strategic Insight: Inevitability is a comfortable assumption for a buyer as well as for a vendor, because it removes the need to plan for the alternative. The organisations that will handle the next two years well are the ones treating adoption as something that has to be earned at each permission point, not something that arrives on schedule.
Your next steps
Immediate actions (this week):
- List the permissions your AI programme depends on and name an owner for each
- Ask managers whether any AI objection has been settled informally this year
- Identify which of your published figures a competent opponent could recalculate
Strategic priorities (this quarter):
- Attach a delay cost, not a probability, to each permission failing
- Move one consultation ahead of the decision it relates to and record the result
- Add annotation and extraction layers to supplier due diligence questions
Long-term considerations (this year):
- Report objections and challenges alongside adoption in programme reporting
- Test whether your consent strategy survives a nationally coordinated objection
- Review whether your growth case still assumes a public that the polling does not show
Source: The World Is Already Resisting AI. Now, There is a List to Prove It. by Petra Molnar (Tech Policy Press, 21 May 2026). UK figures are drawn from prior Resultsense reporting, which cites Ofgem via the BBC and the Ada Lovelace Institute polling of 1,928 UK adults.
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will keep watching whether British consent processes start recording opposition as a scheduling input rather than a communications one. Read more analysis at Insights, or get in touch.