Theory of Change

What we want

Our proposal describes what we want: a global halt to frontier AI development until we know how to do it safely and under democratic control.

This page explains how we plan to get there.

The gap we fill

Good policy proposals for AI governance already exist. Public concern about AI is high and growing. But concern does not automatically become political action. Left unfocused, it dissipates, or gets captured by movements that don’t address frontier risk.

The missing link is an organised political constituency that converts growing public concern into political pressure, making the governance proposals already on the table politically viable. PauseAI Global builds that constituency. We complement technical policy organisations who design specific proposals, by building the movement that demands their adoption. Together, we aim to hand decision-makers everything they need to act: a policy framework ready to implement and a democratic mandate that makes acting politically advantageous.

How activation actually works

Effective organisations are communicating AI risks, but awareness alone does not produce political action. Social movement research shows that behaviour change requires community reinforcement: people activate when their peers activate, not by reading articles.

PauseAI is the activation layer. We train organisers who build local teams and train new organisers within their own communities. This is a chain reaction model designed to compound. PauseCon conferences equip people with organising skills. Protests provide the activation energy that tips someone from concerned to committed. Our federation model enables national chapters to replicate the playbook independently.

Two conditions must hold for the chain reaction to fire:

1. People must understand the severity of the situation: that the race to build AGI poses catastrophic and existential risks to them and their families.
2. People must believe action is possible, that a pause is achievable, not utopian.

The second is as important as the first. A lack of hope produces despair and paralysis, not movement. This is why we organise around a concrete, unifying demand: an international pause on frontier AI development. It is specific enough to mobilise around, broad enough to unite people with different concerns, and high-optionality by design. It directly counters the most powerful obstacle we face: the widespread, self-fulfilling belief that AI development simply cannot be stopped.

This is also why our communications discipline pairs the severity of the risk with a concrete call to action. Join us, contact your representative, back this proposal. We never leave people with alarm and no remedy.

What it takes to get a pause

A pause on frontier AI development is an international agreement. International agreements of this kind require five conditions to hold simultaneously:

1. Political will in multiple major-power governments sufficient that defection is unattractive.
2. A technical and governance framework ready to execute.
3. A triggering window, almost certainly event-driven, sufficient to break political inertia.
4. Elite consensus that pausing is the responsible position, separating expert defenders of the industry from its commercial defenders.
5. A public mandate visible enough to make signing politically rewarding rather than costly for the leaders who do it.

PauseAI Global’s primary lane covers conditions 1, 3 and 5: building the coordinated multi-jurisdiction public mandate, structuring the capacity to convert warning-shot windows into political momentum, and making the politics of signing favourable for governments that would act.

Conditions 2 and 4 are the lane of policy organisations like MIRI, FLI and ControlAI. They develop the proposals, brief decision-makers, and build expert consensus, while we build the political conditions under which proposals get adopted.

PauseAI cannot deliver a pause unilaterally. PauseAI Global can build the substrate that makes a pause possible when the other conditions are met.

Why a federation, specifically

There are three plausible mechanisms by which public mandate could convert into a treaty:

1. Treaty convener. A country with the diplomatic standing to call an international process, most plausibly the UK given its Bletchley precedent.
2. Demonstration cascade. One jurisdiction passes meaningful frontier restrictions and the precedent forces others to respond.
3. Multi-jurisdiction simultaneous pressure. Coordinated domestic pressure in several major countries at once, making the political economics of international agreement favourable to all of them at the same moment.

The federation model is designed around the third mechanism, which we believe is the most likely to produce durable outcomes. A single-country organisation, however effective, cannot produce multi-jurisdiction simultaneous pressure. A coalition of national chapters running coordinated campaigns on shared timelines, with shared messaging and shared infrastructure, can.

This is the specific structural claim that distinguishes our approach from country-level advocacy. The infrastructure we are now setting up, namely the Federation Charter  , shared training pipelines, coordinated lobbying tools  , and rapid response protocols, is what makes this repeatable at scale.

Why now

Frontier labs are now using their own models to build the next generation, and there are credible reasons to believe recursive self-improvement could cross a critical threshold within the next 12 months. AI-driven economic disruption is accelerating and public anxiety is growing fast.

Without a disciplined, catastrophic-risk-focused movement ready to channel that energy, it will either dissipate or default into incoherent backlash. We are currently the only global grassroots organisation positioned to fill that gap. The infrastructure must exist before major disruptions hit, not after. We are embedding in affected communities now: students facing a transformed job market, workers facing displacement, democracy advocates. We build relationships and credibility so that when disruptions intensify, we are already trusted voices with the organisational capacity to convert unfolding events into political momentum.

What we do

Our activities serve three functions: reaching new audiences, activating them into organisers who replicate the model, and converting the resulting constituency into political pressure.

Train organisers. PauseCon conferences are our primary capacity-building tool. PauseCon Brussels (February 2026) trained ~80 organisers from 15 countries in leadership, campaign strategy and communications. Participants are already delivering independently: new chapters launched, meetings with parliamentarians secured, campaigns running. We run two PauseCons per year, each producing a cohort of trained organisers who return to their countries equipped to grow the movement locally.

Scale the federation. We provide standardised support infrastructure for national chapters: organising playbooks, fundraising guidance, coordinated lobbying tools, and seed funding for new chapters. A tiered model (Local group → Chapter → Independent entity) provides differentiated support based on chapter maturity, while non-negotiable standards (core values, mission, core positions  , messaging guidelines  , code of conduct) protect the brand and the movement’s integrity.

Run coordinated campaigns. Quarterly campaigns aligned across all chapters on shared themes and timelines. Each campaign is a strategic experiment testing what resonates, which communities mobilise, and what creates political pressure. Results inform the next one. The India Summit campaign in February 2026 generated roughly 2,000 targeted emails to representatives across 14 countries and a petition with over 4,000 signatures, while testing our ability to coordinate the federation at scale.

Build coalitions. We build two types of partnerships. Movement-building coalitions give us access to communities already affected by AI: student unions, labour organisations, parent groups, democracy and rights advocates. Policy partnerships with organisations like MIRI and ControlAI ensure we advocate for rigorous, well-designed proposals and give us somewhere to route politicians who are ready for detailed briefings.

Apply political pressure. Grassroots lobbying across 15+ countries, tiered by volunteer commitment. Anyone can send a personalised message to their representative in under a minute using our coordinated lobbying tool. More dedicated volunteers seek in-person meetings with their local representatives. Senior staff pursue high-profile meetings and route them toward trusted policy partners for detailed briefings. Each year includes at least two coordinated global protests to signal constituency size to politicians and media.

Maintain public visibility. CEO media appearances, conference speaking, articles, press releases, video content and a sustained social media presence position PauseAI as a recognised voice of public concern about AGI. A warning shot activation protocol enables us to convert unfolding events into political moments within hours rather than weeks.

What we don’t do

We don’t tolerate violence or violent rhetoric. We are a peaceful movement and make this clear in our Protestor’s Code of Conduct, Discord rules, and Volunteer agreement. We are the good guys, and we want the public to be on our side.

We don’t take sides on unrelated issues. We are a movement focused on an AI pause. We do not take positions on topics outside our remit, even when short-term opportunities make it tempting.

We avoid dishonesty. We need people to trust what we say. Honesty is non-negotiable, including about our own uncertainties and limitations.

Key uncertainties

We are honest that several things could go wrong, and we manage them actively rather than ignoring them.

Federation coordination. Our model depends on chapters aligning on shared campaigns and messaging while retaining local autonomy. The Federation Charter is the structural mitigation, and its enforcement is tested with every coordinated campaign.

Capture by adjacent movements. Public concern about AI is currently diffuse and could consolidate around frames that don’t address frontier risk. Our mitigation is messaging discipline and explicit commitment to the frontier-development frame.

Warning shot judgement. The rapid response protocol depends on calls about which events merit activation. Activating for an event that doesn’t land risks crying wolf; failing to activate for one that does risks missing the window.

Measurement. Several of our success metrics (media reach, protest attendance, petition signatures) are proxies rather than direct measures of pause probability. Building better measurement is an active priority.

Get involved

The movement has hundreds of new prospective volunteers signing up every month. Our bottleneck is the capacity to train, support and empower them.

Join PauseAI and take action.