Holly Elmore

Holly Elmore is the founder and Executive Director of PauseAI US — the US national chapter Elmore founded in 2023 and now runs as an independent US nonprofit with its own staff and donor base distinct from PauseAI Global under Joep Meindertsma and Maxime Fournes (see Person entry) — and is the corpus's on-record US voice on the case for a treaty-backed pause on training the most powerful general AI systems, the substantive case that uncontrolled superintelligence is best understood through the lens of evolutionary biology and ecological displacement, and the named critique of safety-researchers being captured by the AGI companies they hoped to reform. She is tracked here as a Voice because her sustained named-byline public output — the 9 December 2025 UC Berkeley AI Risk Speaker Series talk anchored on the Radium Girls analogy and the "species go extinct as a rule" frame; the FOOM Magazine December 2025 Q&A with Mordechai Rorvig anchored on the evolutionary-biology-as-prior-for-AI-risk framing; the January 2024 Inside View podcast with Michaël Trazzi anchoring the indefinite-global-pause framing; the November 2024 EA Forum AMA anchoring the Overton-window and radical-flank theory-of-change framings; the 21 March 2026 Stop the AI Race speech in San Francisco at what PauseAI US describes as the largest AI-safety protest in US history; the Fortune April 2026 named on-record commentary distinguishing PauseAI US from Stop AI after the alleged attack on Sam Altman's residence; the recurring hollyelmore.substack.com Substack newsletter; and her sustained EA Forum essay register — carries the working argument that frontier-AI training above a defined capability or compute threshold is best paused now by treaty and policy-level intervention rather than by alignment-research alone, that the substantive case for pause is owed to a public-organiser register engaging non-AI audiences in US grassroots protest and Congressional lobby rather than to insider AI-safety conversations alone, and that the central capture risk inside the AI-safety field is the structural capture of safety researchers by AGI companies.

The Voice anchors three movement-area registers that the corpus's voices slice had previously left empty or under-tended.

• The first US pause-movement voice anchor. The corpus's existing pause-movement voice anchors run through the Netherlands (Joep Meindertsma, software-entrepreneur-turned-organiser, PauseAI Global Founder and Board Chair) and France (Maxime Fournes, machine-learning-engineer-turned-organiser, PauseAI Global CEO since December 2025) — leaving the US national-chapter voice anchor untouched despite PauseAI US running as an independent US nonprofit with its own staff and donor base, despite its named Stop the AI Race March in San Francisco on 21 March 2026 framed by the organisation as "the largest AI safety protest in US history (so far!)", and despite its named Congressional-lobby register on the bipartisan AI Risk Evaluation Act and the Day in DC representative-meeting programme. This Voice anchors the US pause-movement voice through a US national-chapter independent of PauseAI Global, complementing the matched pair of pause-movement leadership anchors the corpus already tracks through Meindertsma and Fournes.
• The evolutionary-biologist-turned-organiser sub-type. Structurally distinct from the corpus's existing pause-movement voice anchors — Meindertsma's software-entrepreneur-turned-organiser register rooted in the Dutch civic-tech field, Fournes's machine-learning-engineer-turned-organiser register rooted in the Cambridge mathematics and London chief-data-scientist field — Elmore's distinctive register is the Harvard-doctorate-in-evolutionary-biology-turned-organiser whose prior research career under David Haig and three-year wild-animal-welfare research register at Rethink Priorities sits beside her PauseAI US founding and her Harvard EA-organizing register (2014–2020). The evolutionary-biology side anchors the substantive case that uncontrolled superintelligence is best understood through the lens of ecological displacement and species extinction — "the only thing that would prime you for understanding that is having an empirical knowledge of evolution" — and the named "species go extinct as a rule" framing of the Berkeley Talks register; the EA-organizing side anchors the structural-capture critique of safety researchers joining AGI companies; the wild-animal-welfare side anchors the cross-cutting framing she carried from her FLI AI Safety Community Researchers bio that "existential safety is every person and every animal's safety".
• The Overton-window and radical-flank theory-of-change register. The named theory-of-change framing Elmore carries through her EA Forum AMA and her FOOM Magazine Q&A — that the most positive impact of PauseAI's grassroots register comes from shifting public perception toward accepting moderate AI safety measures rather than from achieving the pause treaty itself, and that PauseAI's clearer messaging and grassroots approach complements more constrained lobbying from aligned organisations by making their asks seem safer — anchors the substantive radical-flank theory-of-change register the corpus's voices slice had previously left untouched on the AI-safety side. The framing carries directly through into the substantive case made in Elmore's named-byline EA Forum essay register on capture-of-safety-researchers, the case against waiting-on-alignment, and the case against "surgical pause" timing-based approaches.

Public output and venues

Elmore's public-facing work runs through four overlapping channels.

• The named US-academic-and-Berkeley public-speaking register. The headline anchor of Elmore's US-audience public-speaking register is the 9 December 2025 Berkeley Talks delivered through UC Berkeley's AI Risk Speaker Series and published as a podcast episode by UC Berkeley News on 6 February 2026. The talk anchors the substantive Radium Girls analogy ("we're repeating this mistake with artificial intelligence by assuming we can safely play with a technology we don't fully understand"), the "species go extinct as a rule" framing of the fragile-world view inside which the pause case sits, and the substantive rejection of the position that AI-alignment alone can carry the load of safety ("the expectation of many people in AI safety, for many years, has been that when we got to this point, the AI, once it was aligned, would figure out the answers for us"). The Berkeley Talks register places Elmore alongside the US academic-and-civic AI-safety public-speaking field, and corroborates her named co-founder relationship to Meindertsma in the wider PauseAI Movement.
• The named long-form podcast and Q&A register on the case for pause. Elmore's sustained podcast-interview register runs through The Inside View with Michaël Trazzi (19 January 2024) anchoring the indefinite-global-pause framing and the named warning against safety-researchers joining AGI companies ("People who care about AI safety...shouldn't work with the AGI companies"; "we don't know how to time [a surgical pause] and I think that idea is just a gift to people who want to get in the way"); through the FOOM Magazine December 2025 Q&A with Mordechai Rorvig anchoring the evolutionary-biology-as-prior-for-AI-risk framing and the named Overton-window framing of PauseAI's theory of change; and through the 25 January 2026 YouTube long-form interview "The Deep Worldview and Theory of Change Behind PauseAI, from a Founder" anchoring the named-byline interview register through which her substantive worldview on genetic conflict and social change carries into the wider AI-safety podcast-and-YouTube field. The podcast-and-Q&A register anchors the substantive policy-and-strategy register that the named-byline EA Forum essay register elaborates in long-form text.
• The named US-grassroots-mobilisation and rally register. The most visible US-civil-society-audience-facing surface of Elmore's voice is the 21 March 2026 Stop the AI Race March in San Francisco — the rally PauseAI US's own site frames as "the largest AI safety protest in US history (so far!)" — where Elmore delivered the named "YOU have the power" speech anchoring the substantive framing that "we AREN'T marching to beg the AI companies to be merciful and spare our lives", the conditional-pause-mechanism framing (companies should voluntarily agree to pause if all frontier labs agree), the named critique of Anthropic citing competitive concerns as a barrier to conditional pause commitments, and the "each act of courage kindles more courage" framing. The mass-mobilisation register sits alongside the PauseAI US Congressional-lobby register that runs through the bipartisan AI Risk Evaluation Act advocacy, the Day in DC programme, and the named local-group lobby meetings with Senators Murkowski and Alsobrooks and Representative Lieu — anchoring the inside-outside-game pair that PauseAI US carries on the US national surface.
• The named EA Forum, Substack, and US press register. Elmore's named-byline written-output register runs through the EA Forum essay register — a sustained 56-post, 7,681-karma register including the curated post "The myth of AI 'warning shots' as cavalry" and the named posts "Selling out to AI companies is bad. Period. You will be corrupted.", "Influence from within on AI doesn't work, so bye bye", "Enough about AI timelines — we already know what we need to know", "Genes did misalignment first: comparing gradient hacking and meiotic drive", and "Thoughts on my relationship to EA (and please donate to PauseAI US)" — and through the hollyelmore.substack.com Substack newsletter. The US press register runs most visibly through the Fortune April 2026 named on-record commentary distinguishing PauseAI US from Stop AI after the alleged attack on Sam Altman's residence — where Elmore's named framing of PauseAI US as the vehicle for "democratic change through democratic means" anchors the on-record positioning that the CNN April 2026 reporting and the parallel European-press register on the same incident corroborate from Fournes's CEO surface.

Signature framings

Three formulations recur across Elmore's public output and have done the most to install her register into the US AI-safety, pause-movement, and EA-to-PauseAI fields.

• Evolutionary biology as a prior on AI risk — the species-extinction framing. Elmore's framing in the FOOM Magazine Q&A and the Berkeley Talks — that "the only thing that would prime you for understanding [AI risk] is having an empirical knowledge of evolution", that "species go extinct as a rule" and humans remain vulnerable to disrupting environmental, social, and institutional equilibria they do not fully comprehend, and that the Radium Girls tragedy is the working analogy for societies unknowingly exposing themselves to catastrophic technological risk — is the most condensed single articulation of the evolutionary-biology-as-prior-on-AI-risk case the corpus's pause-movement voice register turns on. The framing displaces the engineering-and-alignment-only framing of the US AI-safety conversation with a substantive ecological-and-evolutionary framing rooted in Elmore's Harvard doctorate under David Haig, and anchors the case that the AI-safety question is at its core a fragile-world question rather than a technical-ML-credentials question.
• "Selling out to AI companies is bad. Period." — the capture-of-safety-researchers framing. Elmore's named EA Forum post "Selling out to AI companies is bad. Period. You will be corrupted." (and the parallel framing in the Inside View interview that "people who care about AI safety...shouldn't work with the AGI companies") anchors the substantive critique that safety researchers joining the AGI companies they hope to reform are structurally captured by them — "Everything is stacked in its favor" once the researcher is inside the organisation, and the central question becomes "how are you equipped to resist it changing you?". The framing carries directly through into PauseAI US's strategic positioning as a public-organiser surface outside the AGI-company-and-AI-safety-think-tank ecosystem and into the EA Forum essay register on capture, influence, and the limits of inside-game AI-safety work.
• Overton-window and radical-flank theory-of-change framing. Elmore's named theory-of-change framing in the EA Forum AMA — that the most positive impact of PauseAI's grassroots register comes from shifting public perception toward accepting moderate AI safety measures, that PauseAI's clearer messaging and grassroots approach functions as a radical flank making aligned organisations' more constrained asks seem safer, and that pushing the pause idea "further into the center of the Overton window" eventually makes it "what we do as a society" — anchors the substantive radical-flank theory of change the US AI-safety-and-pause-movement field had previously left under-articulated. The framing carries directly through into PauseAI US's organising posture of public protest, mass petition, and sustained Congressional-lobby engagement on the bipartisan AI Risk Evaluation Act, and into Elmore's named-byline critiques of "warning shots", AI-timelines-as-organising-time, and the "surgical pause" idea on the EA Forum.

Why this is a Voice entry

A Voice entry is created here, rather than additional structure on the Person entry, because Elmore's public-facing output is itself the load-bearing object the corpus needs to track: the UC Berkeley AI Risk Speaker Series Berkeley Talks appearance; the FOOM Magazine December 2025 Q&A with Mordechai Rorvig; the Inside View podcast with Michaël Trazzi; the named EA Forum AMA; the 21 March 2026 Stop the AI Race speech; the Fortune April 2026 commentary; the recurring Substack newsletter; and the sustained EA Forum essay register including "The myth of AI 'warning shots' as cavalry", "Selling out to AI companies is bad. Period.", "Influence from within on AI doesn't work, so bye bye", "Enough about AI timelines — we already know what we need to know", and "Genes did misalignment first". The corpus's pause-movement voice slots previously carried Meindertsma's Netherlands software-entrepreneur-turned-organiser register and Fournes's France machine-learning-engineer-turned-organiser register; this entry gives the corpus its first US pause-movement voice anchor, the evolutionary-biologist-turned-organiser sub-type, and the Overton-window-and-radical-flank theory-of-change register that previously sat untouched on the AI-safety side of the corpus's voices slice. Affiliation, training, and biographical detail are recorded on the linked Person entry per the corpus's Person/Voice split.