PauseAI international protests, open letters, and parliamentary lobbying (2023–ongoing)

Since May 2023, PauseAI — the international grassroots movement founded by Joep Meindertsma in Utrecht — has run a continuous, multi-year public-mobilization campaign outside the offices of the major frontier-AI labs and around the international AI Safety Summit series. The campaign's stated goal is a coordinated, treaty-backed pause on the training of general AI systems above a defined capability or compute threshold; its working theory is that visible, peaceful, mass-political pressure on governments and labs is a precondition for the kind of binding international agreement the policy track alone has so far been unable to produce. By mid-2025 the campaign had accumulated more than thirty separately staged protests in at least twenty countries, an open-letter and constituency-lobbying track culminating in a 60-politician August 2025 letter to Google DeepMind, and a working repertoire of named-target actions, coordinated days of action, and street-theatre tactics that distinguishes PauseAI from the civil-disobedience-focused Stop AI and the Westminster-inside-game ControlAI in the comparative landscape of anti-frontier-AI organizing.

Origins (May–November 2023): from Microsoft Brussels to Bletchley Park

PauseAI's first public action was a small protest outside Microsoft's Brussels lobbying office on 23 May 2023, staged during an AI policy event in the European quarter and assembled within weeks of the organisation's founding. Through the rest of 2023 the campaign ran a steady local-action cadence — pickets and rallies in London (Parliament Square, the UK Office for AI, the Department for Science, Innovation and Technology), Melbourne, New York, San Francisco, and The Hague — and by 21 October 2023 had organized its first multi-city day of action, simultaneous demonstrations across San Francisco, Boston, London, Den Haag, Melbourne, Ottawa, Copenhagen, and Jerusalem. The October action issued a three-part set of demands on policymakers (pre-training regulation and hardware restrictions), companies (publicly support pausing development), and summit attendees (prioritise safety over economic growth), and was explicitly framed as a curtain-raiser for the November summit.

On 1–2 November 2023 PauseAI organized a protest at the inaugural AI Safety Summit at Bletchley Park, the campaign's first action targeting an intergovernmental AI summit. Meindertsma described the resulting Bletchley Declaration as "a small first step" but argued in his on-record statements that durable safety would require binding international treaties on the model of the Montreal Protocol and the laser-weapons ban — the framing that the campaign has since carried through Seoul, Paris, and the follow-on conferences in 2024 and 2025.

The OpenAI militarisation protest and the May 2024 day of action

On 12 February 2024, PauseAI US — by then under the direction of Holly Elmore — assembled outside OpenAI's San Francisco headquarters in a coalition action with the smaller, separate No AGI group represented by Sam Kirchner. The proximate trigger was OpenAI's 10 January 2024 removal of usage-policy language prohibiting military and warfare applications and the reported acceptance of the Pentagon as a client; the protest demanded that the company end its Defence relationships and pause its AGI work. The action was the campaign's first picket framed primarily around a single lab's policy choice rather than around an upcoming summit, and was the proof of concept for the named-target action format that would recur at OpenAI, Anthropic, and Google DeepMind sites through 2024 and 2025.

The 13 May 2024 international day of action, staged ahead of the Seoul AI Summit on 21–22 May, was the campaign's first large-scale coordinated mobilization. Protests ran simultaneously in fourteen cities across seven countries — San Francisco, New York, Portland, Ottawa, Reykjavík, London, Berlin, Stockholm, Den Haag, Paris, Rome, Oslo, Sydney, and São Paulo — and were framed by the campaign as a message to the ministers convening in Seoul to "be the adults in the room" on frontier-AI risks. The campaign followed through with a 20–22 November 2024 international action timed to the San Francisco AI Safety Conference, running pickets across eleven cities (including a stop at Anthropic's San Francisco offices) and calling on "one country to step up and initialise treaty negotiations."

The Paris Summit and the Google DeepMind protest (2025)

The campaign's 7–11 February 2025 mobilization around the Paris AI Action Summit was its largest summit-targeted action to date, with demonstrations in roughly twenty cities — Paris, London, Berlin, Brussels, Zürich, Prague, Milan, Stockholm, Copenhagen, Oslo, Kristiansand, Trondheim, New York, Victoria, Kinshasa, Brazzaville, N'Djaména, Melbourne, and others — and a four-part demand set on the summit's organizers: centre AI safety as the meeting's focus, commit to collaboration on global treaties, establish international enforcement bodies, and share national best practice on AI-safety contributions. Maxime Fournes — who would shortly succeed Meindertsma as global CEO — coordinated the campaign's French-side organizing through PauseIA France.

On 30 June 2025, PauseAI staged a sustained picket and street-theatre action outside Google DeepMind's London headquarters that PauseAI's own post-action report characterized as the largest AI-safety protest then on record. The proximate cause was Google's release of Gemini 2.5 Pro: on PauseAI's reconstructed timeline, the model was made publicly available on 25 March 2025 without a safety report, was characterised on 3 April as "experimental" to avoid the Frontier AI Safety Commitments that Google had signed at Seoul, on 9 April produced no response to questions about UK AI Security Institute involvement, on 16 April was accompanied by a short model card with no detail on external testing, and on 28 April acquired a vague reference to "third-party external testers" with no further identifying detail. Around five dozen demonstrators marched through King's Cross and assembled outside DeepMind's offices for a gavel-banging mock trial with an improvised jury, chanting "Stop the race, it's unsafe" and "Test, don't guess"; PauseAI organising director Ella Hughes framed the situation publicly with the line that "right now, AI companies are less regulated than sandwich shops." Sitting MP Iqbal Mohamed joined the action.

The open-letter and parliamentary track

The DeepMind protest was followed by a sustained constituency-lobbying push and an open letter published on 29 August 2025 addressed to Google DeepMind. PauseAI coordinated the letter; it was co-signed by 60 UK politicians from more than 10 parties — including Lord Browne of Ladyton (Labour, former Defence Secretary), Baroness Kidron (Crossbench), Lord McNally (Liberal Democrat), Baroness Chakrabarti (Labour), Desmond Swayne MP (Conservative), all four Green Party MPs, and all four Plaid Cymru MPs — together with civil-society partners Open Rights Group, Connected by Data, the Safe AI for Children Alliance, and Open Data Manchester. The letter demanded that Google define "deployment" in line with common understanding (so that "experimental" availability could not be used to avoid the Seoul commitments), publish safety-evaluation reports on a fixed timeline for every future frontier-model release, and disclose which government agencies and independent third parties had tested its models, with exact timelines. Fortune's coverage framed the letter as a parliamentary "breach of trust" accusation; a Google DeepMind spokesperson responded that its models "undergo rigorous safety checks, including by U.K. AISI and other third-party testers — and Gemini 2.5 is no exception." The letter is the most prominent legislative-side artefact the campaign has produced to date and the clearest illustration of the campaign's working theory: that visible street protest is one stage in a longer arc that runs through constituency offices, signed letters, and named-counterparty accountability rather than to a one-off media moment.

Tactics and posture

Across three years, the campaign has settled into a recognisable tactical repertoire — coordinated international "days of action" timed to the AI Safety Summit series, named-target pickets at specific lab sites (Microsoft Brussels, OpenAI SF, Anthropic SF, Google DeepMind London), street-theatre staging (the King's Cross mock trial; gavel-and-jury formats; signage built around durable framings such as "When in doubt, pause" and "Test, don't guess"), open letters co-signed by serving parliamentarians and named civil-society partners, and an explicit posture of peaceful, lawful, mass-political organising. The campaign's comparative positioning within the wider AI-safety advocacy space is built on that posture: where Stop AI pursues civil-disobedience tactics (lock-ons, building blockades) and ControlAI pursues a Westminster inside-game, PauseAI runs the street-protest-and-public-letter lane and has been the most visible occupant of that lane since 2023.

Significance for the broader AI-good movement

For the wider make-AI-good movement this corpus is mapping, the campaign matters on three connected counts. First, it is the longest sustained street-protest record in the AI-safety field on the public record — a multi-year, multi-continent demonstration that visible, lawful, mass-political organising on frontier-AI training is not only possible but reproducible across national contexts, from the European capitals to North American tech hubs to São Paulo, Sydney, and central Africa. Second, by tying the protest track to a calendar of intergovernmental summits and to a sequence of named-lab grievances (OpenAI's military-use policy reversal, Google DeepMind's handling of Gemini 2.5 Pro), the campaign has produced a working template — visible action plus named-target accountability plus signed parliamentary letter — that subsequent grassroots AI-governance organising can cite as proof of concept. Third, the campaign has converted a small set of durable framings ("AI companies are less regulated than sandwich shops"; "we need binding international treaties on the model of the Montreal Protocol"; "test, don't guess") into widely circulated public artefacts that have travelled well beyond the protest line — into Fortune, Wired, TIME, the Financial Times, and the floor of Westminster — and that future entries on the campaign's likely successor mobilizations around the 2026 summits and follow-on lab releases will continue to draw on.