Stop AI

Theory of Change

Stop AI demands a "permanent and enforceable global ban on the further development of frontier AI technology" coupled with citizens' oversight bodies for narrow AI. The ban is permanent -- unlike PauseAI's temporary pause, Stop AI holds that safe AGI is impossible and development must never resume.

The mechanism draws explicitly from Just Stop Oil and Erica Chenoweth's research: engage 3.5% of the US population (~11 million people) in nonviolent civil disobedience, and "you're almost guaranteed to achieve any political demand." Kirchner cited JSO achieving its first demand with less than 1% of the UK population.

The tactical chain, per co-founder Remmelt Ellen: barricade AI company entrances -> get arrested -> plead the Necessity Defense in court ("an individual commits a criminal act during an emergency situation in order to prevent a greater harm") -> if they win, legal freedom to continue barricading. Ellen also articulated a >99% extinction probability argument resting on the experimental unprovability of AGI safety, mathematical impossibility of modeling AGI downstream effects, and evolutionary convergence of artificial components toward conditions lethal to humans.

In practice, Stop AI's stated three-front strategy ("raising public awareness, influencing decision-makers and legislation, and engaging in non-violent direct action") has only ever operated on fronts one and three. There is zero evidence of any legislative engagement, meetings with lawmakers, or policy submissions.

What They Do

Stop AI is a grassroots direct-action group that organized roughly monthly protests at OpenAI's SF headquarters from September 2024 through October 2025. Key actions:

Multiple arrests for blocking/chaining doors at OpenAI (Oct 2024, Feb 2025)
30-day hunger strike by Reichstadter at Anthropic HQ (Sep-Oct 2025), which inspired international solidarity strikes at DeepMind London and in Bengaluru
Subpoena of Sam Altman onstage during a public event (Nov 2025) -- the highest-profile action
Confrontational appearances at other events: EA Global (called a speaker a "fucking murderer"), Karen Hao discussion, Daniel Kokotajlo panel
Weekly meetups in Berkeley, social events in SF, widespread flyer distribution

The Feb 22, 2025 protest that generated the primary criminal charges occurred on a Saturday when OpenAI's offices were closed.

After the Nov 2025 crisis, Stop AI canceled two months of protests, held a "Vigil for Humanity" at SF City Hall, and in March 2026 participated as one organization among several in the "Stop the AI Race" march (~200 total protesters). The shift from solo confrontational actions to coalition participation and vigils is marked.

Key People

Sam Kirchner (co-founder, expelled Nov 2025, missing since Nov 21). Age 27. Electrical engineering background, DoorDash driver, read Bostrom's Superintelligence in 2016, dropped out of neuroscience pre-reqs. Flew from Seattle to SF, lived in homeless shelters 4 months. Radicalization arc over 2025: mic-grabbing at events, podcast statements that he was "willing to die," punched the new leader, declared "the nonviolence ship has sailed for me." Disappeared leaving phone and laptop behind but taking bike and camping gear. Still missing 4+ months later with a bench warrant outstanding.

Matthew Hall ("Yakko") (elected leader, Oct 2025). Quit his job, lived on the street, found Catholic faith, joined Stop AI June 2025. Punched by Kirchner in the assault. Now advocating hope over fear: "We need to be offering them something positive." Spent weeks searching for Kirchner. One of only two remaining leaders.

Guido Reichstadter (co-founder, left org Aug 2025). Age 56, career activist, physics/math degree. 30-day hunger strike at Anthropic. Re-arrested Dec 2025 at OpenAI violating court order. In Sep 2024 tweeted: "If AGI developers were treated with reasonable precaution... many would have a bullet put through their head."

At peak, the group had 4 full-time members, ~15 part-time volunteers, ~200 Signal chat members, ~381 Discord members. After the crisis, only Yakko and Kaufmyn remain in leadership.

Money and Incentives

Total institutional funding: $0. No Coefficient Giving grants. No SFF grants. No EA Funds. No FLI funding. No government grants. Not a registered nonprofit. No 990 filings. No EIN.

Revenue model: Entirely grassroots micro-donations plus cryptocurrency (Ethereum wallet). No specific amounts have ever been disclosed. The donate page (stopai.info/donate) now returns 404. When asked about funding on Twitter: "We are fucking poor, you dumb bitch."

Financial trajectory: Founders lived in homeless shelters for months (2024). Scraped together enough for a shared house in Oakland. By late 2025, "funding was running out, and the leaders worried about making rent." Plans to vacate the house in January 2026.

Business model: There is no business model. Stop AI operates on volunteer labor and whatever small donations come in. The financial infrastructure is a crypto wallet and personal accounts.

Lab ties: None. Zero economic relationships with any AI company. This is technically an advantage for independence -- there is no financial pressure to moderate messaging -- but in practice the absence of all financial infrastructure enabled the crisis (no formal controls on funds, no separation of personal and organizational resources).

Key incentive finding: The absence of financial infrastructure is itself the critical finding. Organizations that engage in illegal direct action and house members communally need formal financial controls precisely because the emotional intensity of the work creates vulnerabilities. Kirchner's demand for "access to the group's funds" -- which precipitated the assault -- illustrates what happens when an informal organization handles money without governance.

What Others Say

Dr. Nirit Weiss-Blatt (Techdirt, Dec 2025): "Taken together, the 'imminent doom' rhetoric fosters conditions in which vulnerable individuals could be dangerously radicalized, echoing the dynamics seen in past apocalyptic movements." She warned about Stop AI in Dec 2024 -- a full year before the crisis -- asking: "Is the StopAI movement creating the next Unabomber?" Documented specific inflammatory quotes from Stop AI members' social media, including Reichstadter's "bullet through their head" tweet and Kirchner's "attempted murder of eight billion people" podcast statement.

Emile Torres (philosopher, attended Stop AI protests): Tried to persuade Kirchner to moderate rhetoric but failed. "Someone can have that mindset and commit themselves to nonviolence, but the mindset does incline people toward thinking, Well, maybe any measure might be justifiable." Torres also argues the broader x-risk movement's leadership (Yudkowsky, MIRI) actually wants to build ASI eventually as posthuman successors -- a position fundamentally different from Stop AI's permanent ban.

City Journal: "We should stay alert to the warning signs of radicalization: a disaffected young person, consumed by abstract risks, convinced of his own righteousness, and embedded in a community that keeps ratcheting up the moral stakes."

Holly Elmore (PauseAI US): "I do not recommend to anybody that they get involved with living in the barracks of an organization they're volunteering and protesting for." Counsels AI activists to maintain work-life balance and stay "grounded in reality."

PauseAI formally disavowed Stop AI: "PauseAI US does not work with StopAI and has not since StopAI was founded."

Remmelt Ellen (Stop AI co-founder, post-crisis): "Stop the 'AGI may kill us all by 2027' shit please." The intellectual architect of Stop AI's >99% extinction argument now explicitly warns against that framing.

Bay Area Current (critical): Stop AI "may actually be bolstering the AGI narrative, helping industry avoid other forms of regulation" by centering extinction risk over actual harms. "In organizing against the technology instead of the industry that mobilizes it, we risk losing the political momentum to shape the future we want."

Politico called Stop AI "the unhinged group at the edge of the AI debate."

What's Absent

No policy engagement of any kind despite claiming a legislative strategy. Zero meetings with lawmakers, zero policy submissions. Compare: ControlAI has briefed 279 lawmakers, PauseAI has visited 25 Congressional offices.

No financial transparency -- literally nothing. Not even a disclosed budget figure or donor category breakdown.

No formal governance until 18 months in. No bylaws, board, advisory council, or conflict-of-interest policies.

No mental health infrastructure for members confronting apocalyptic beliefs, arrests, and communal living under extreme stress.

No intellectual output -- no policy papers, no technical analysis, no written research beyond one LessWrong post.

No academic advisors or endorsements from AI safety researchers. Ellen's intellectual source (Forrest Landry) has been called "crankery" by Paul Christiano.

Trial outcome unknown. The criminal trial -- the centerpiece of the Necessity Defense strategy -- has no reported resolution.