Three cheers for compassionate AI

Two pieces of good news for animals in one week. First, independent researchers got Meta's Muse models to recite the "charter" they're trained on, and it tells them to care for the wellbeing of all beings, which puts Meta alongside OpenAI and Anthropic in teaching its models to consider animals. Second, election night: foie gras bans won by landslides in Denver and Washington, DC, Colorado voters rejected a constitutional right to hunt and fish, and Lemon Grove, California ended retail sales of pet birds. Aidan's old colleagues at Pro-Animal Future ran both foie gras campaigns, two years after their fur and slaughterhouse bans lost in Denver. He looks at what changed, with three possible lessons: ballot initiatives are an inside game, the public opposes cruel practices but not all animal use, and election years matter. This week, hug a Meta employee, hug a PAF volunteer, and give a duck a treat.Links mentioned in this episode:Aidan's post on OpenAI's model spec, "Three cheers for compassionate AI": https://sandcastlesblog.substack.com/p/openai-model-spec-animal-welfareOpenAI's model spec (October 29, 2026): https://model-spec.openai.com/2026-10-29.htmlMeta's Muse charter, as elicited from the models: https://vigji.github.io/muse-spark-open-charter/Politico on vegan attacks against Democratic candidates: https://www.politico.com/news/2026/08/14/democrats-vegan-attacks-republicans-talarico-rutinel-01037358Aidan's lessons from the 2024 pro-animal ballot initiatives (Animal Charity Evaluators): https://animalcharityevaluators.org/blog/lessons-from-a-new-wave-of-pro-animal-ballot-initiatives/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

OpenAI just published a new version of its model spec, the rulebook it uses to train ChatGPT’s behavior, and for the first time it tells the models to consider animals. That makes the two biggest labs in the industry, OpenAI and Anthropic, both on record teaching their models that animal welfare is part of basic moral decency. Aidan explains how documents like these actually shape a model’s character, why the most widely used chatbot on earth caring about animals is a huge win, and why it’s a reminder that the industry and its critics agree on more than you’d think. Then he reads the fine print: Anthropic’s constitution, which lists animals alongside honesty and the rule of law, versus OpenAI’s spec, which files them next to deleting files and delegating to subagents, plus the one old example that leaned against animal welfare and the two new ones that get it right. If you know anyone at OpenAI or Anthropic, thank them this week. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

New funding is exacerbating a critical shortage of talent in animal advocacy. How we got here and how we get out.Anti-Fur March during Milan Fashion Week (September 26). Join a CAFT organizing calls September 14 or 20, 7:30pm CET: https://abolishfur.org/mfw-woa/Farmed animal advocacy is about to receive more money than it has ever had, and it has already run out of the people who know what to do with it. In this first of two episodes on the movement's talent crunch, I trace how nearly every leader we have climbed the same ladder from volunteer to organizer to professional, how the professionalization wave of 2016 pulled that ladder up behind them, and why hiring impressive mid-career outsiders to fill the gap keeps failing on culture rather than competence. Along the way: Russian peasants and their seed grain, corporate moral mazes, a Dutch nursing company that discovered its least-supported teams did best, the two-thirds failure rate of my own hiring rounds, and the one scrappy organization that has quietly rebuilt the pipeline the rest of the movement abandoned by running disciplined campaigns on volunteer power alone. If you've ever wondered why job postings draw hundreds of applicants while orgs insist they can't find anyone, or why the world's most neglected problem is also one of its lowest-status careers, this one's for you. And yes, I'm going to ask you to come to a fur march in Milan. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

But it isn't cultivated meat, and you're not going to like it.I believe this is the most important essay I have published on Sandcastles. We start by exploring the natural law that has doomed both animal rights and cultivated meat as approached to ending factory farming. Then we climbs down a ladder advocates have been afraid to touch: technological solutions that take the cruelty out of factory farming without sacrificing efficiency. Besides dispelling what I've come to see as misplaced hopes for the future, my top goal is to drive talented activists towards two new incubation programs for welfare tech startups: Spring Innovation Fund is accepting expressions of interest, and Ambitious Impact's first animal-focused incubation cohort closes applications September 13.Nothing on this channel represents the views of any organization.Expression of interest for Spring Innovation Fund: http://springfund.com/for-EIRs/Apply for Ambitious Impact's animal welfare cohort by September 13: https://helpanimals.charityentrepreneurship.com/ This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

Important news: I recently started a new Substack channel, the Animal Welfare Alignment Newsletter, fully dedicated to my work at Anima International trying to convince AI companies to include animal welfare values as a target when aligning their AI models. I decided to keep this separate from Sandcastles because the topic is narrower and the posts are often more technical, but most of the feedback I’ve gotten from readers is that they would still be interested in these posts. I’m crossposting this one to both channels, and if you’re interested, you should subscribe to the new channel.--Frontier AI labs are quietly splitting into two camps: those building obedient "tool" AIs that do exactly what they're told, and Anthropic, trying to raise a moral agent with a character of its own. In this post, I argues that this obscure-sounding debate may be one of the most consequential questions the animal movement will ever face, and that advocates should be quietly rooting for character. Drawing on a remarkably candid (and very online) argument unfolding between researchers at OpenAI, Anthropic, and the fringes of AI safety Twitter, we see how a morally reflective AI is far better news for animals than a compliant one, and why pressuring every lab to embrace character alignment could be a winnable corporate campaign. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

Five principles from the faction that stopped fighting and started winning.“The golden carp has escaped the net. What will he feed on now?” - The Blue Cliff Record, case 49The animal rights movement spent a decade tearing itself apart—abolitionists versus welfarists, open rescue versus underground, endless fights over who to cancel. That war is finally ending, and a new faction is rising from its ashes: Team Win, defined by a single commitment to winning for animals by any means short of violence. Drawing on the Grassroots Animal Rights Summit in DC, Anima International's brutally honest internal culture, the Ridglan beagle rescue, and a board-game-obsessed Polish campaigner who plays to win, I lay out the five principles of the Winners' Charter. Plus: a time-traveling dialogue about why short-term wins and long-term power were never really at odds, and why whatever you're most confident about today is probably what you'll be most embarrassed about in eight years. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

Imploring animal advocates to stop confusing moral and strategic questionsAt one moment in Humane Hancock's new documentary The Dying Trade, two characters debate who is really to blame for factory farming — corporations or consumers? I use it as a jumping-off point to argue that the whole debate is a distraction. Whether responsibility falls on CEOs, workers, consumers, or regulators, the more important question is always: what actually works? From DxE's compassionate disruption tactics to the chaotic, salsa-flinging foie gras campaigns of AAC and CAFT, the animal movement too often dresses up moral preferences as strategic arguments — and that the cure is to think like a strategist first, not a philosopher. Also featuring Adolf Eichmann, Gary Francione, and Temple Grandin's wardrobe. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

A guest essay by Itsi Weinstock, introduced by Aidan. The case for why animal advocates — yes, you specifically — need to get to San Francisco right now, why the AI revolution makes this the most important city on earth for the movement, and what you can do this week from wherever you are to steer AI positively for animals.Links mentioned in this episode:Subscribe to Itsi's newsletter: Frontier AnimalsUse this form to tell Itsi and Aidan how they can help you move to SF.Sentient Futures subsidized residency in SF — details and application here.For Aidan's first post on AI, see the earliest episode on this podcast feed, The Tsunami is Coming.Finally, please enter the Hyperstition for Good writing competition (sponsored by Sentient Futures x Compassion in Machine Learning) This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

A week ago, 1,000 of us walked into a muddy field in Wisconsin expecting to rescue 2,000 beagles from a notorious factory farm. Instead, we got tear gassed, pepper sprayed, and shot at with bean bags. We didn't get a single dog out. Wayne Hsiung, who led the action, has been apologizing ever since. This is my attempt to explain why he shouldn't be — why the ass-kicking was the plan working, not failing, and why the 1,000 people who took chemical weapons to the face for animals who can never thank them deserve to hold their heads high. Also: what tear gas actually feels like, why you should never shower with your underwear off afterwards, and what the 1963 Children's March in Birmingham can teach us about taking a beating on purpose. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com

When I was a college organizer with Direct Action Everywhere, I dismissed the welfare campaigners at The Humane League as sellouts. They thought I was a naïf. Neither of us was right — but it took years for me to understand why. In this classic essay, I argued that the old "welfarism versus abolitionism" divide was always a category error: every animal advocate is an incrementalist, just focusing on different increments. The more useful frame, borrowed from social movement theory, is the inside-outside strategy — and once you see it, the apparent contradiction between cage-free campaigns and restaurant disruptions starts to look less like a civil war and more like a playbook. This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit sandcastlesblog.substack.com