What this project is

We train other organisations around Europe to campaign against autonomous weapons. We created our own method and we called it EduCampaigning: we teach the argument, run workshops where people turn it into their own campaign, then they go and campaign. This grant pays us to build the reusable core of that for autonomous weapons and hand it over to civil-society groups across Europe. Every learning material we produce will be Open Source.

There are three deliverables.

A training package: curriculum, workshop kit, and a briefing that explains why autonomous weapons are the sharp end of AI power concentration, where regulation is stuck, and what we can push our governments to actually do.

A plain-language toolkit: messaging and visuals for people who don't read AI safety papers.

Two or three pilot workshops with partner organisations, to stress-test the materials and train a first cohort.

We've also proposed a $400,000 campaign to Longview Philanthropy (civil society engagement, a public art tour, press work, advocacy at G7, EU Council, UN). If Longview funds it, this Manifund grant becomes the training core that plugs straight in. If they don't, the piece still stands.

Via Lattea is a new non-profit in Rome, working across activism, art, and education from three spaces in Testaccio, Rome. We were founded as the campaigning continuation of Latte Creative, a creative social impact agency with 15+ years of work for Greenpeace International, Amnesty International, Save the Children, Doctors Without Borders, Oxfam, and the Council of Europe. The EduCampaigning model (Education → Workshops → Public Campaigns → Social Change) has been running for years across AI safety, climate, and human rights work.
We're part of the Stop Killer Robots coalition (260+ organisations, 70+ countries).

Theory of impact

The AI catastrophe scenario we care about is lock-in. A small group takes permanent control, and no one can undo it later. Autonomous weapons are where lock-in stops being abstract.

Right now, if you want to hold power by force, you need people. Soldiers, police, officials. Any of them can refuse. That refusal is one of the last real checks we have. Once the killing can be done by machines that don't refuse, the check is gone, and a very small number of people can hold power without anyone's consent. We call this the loyalty bottleneck. Removing it is one of the clearest paths from "powerful AI" to "power that can't be reversed."

You can't fix this with a better-behaved model. You need rules, and rules need public pressure behind them, in the countries that can actually push for them. The political will exists. What's missing is the sustained pressure that connects it to binding action. That's the gap we work in.

This grant sits upstream. Instead of running one campaign, we build the thing that lets many groups run their own — shared materials, a tested workshop model, framing an ordinary person can grasp. Money spent here keeps working after it's gone, because the assets are open and the trained organisations don't need us for daily execution. It's a cheap way to grow the civil-society side of AI safety, which is thin compared to the technical side, and to grow it around the one AI risk that's already loading live ammunition.

How the money will be spent

Minimum $20,000. We build the Open Source reusable core and prove it once. Curriculum and facilitator kit, loyalty-bottleneck messaging and visual toolkit, one pilot workshop with a first cohort of partner organisations, light coordination. Enough to produce assets that any European civil-society group can pick up and use, and to show the model works.

Ideal $120,000. Same core, but as a real programme. Materials professionally produced and translated into four or five languages. Six workshops across EU member states, not one. Small launch stipends so trained organisations actually get their own campaigns off the ground. Nine months of coordination, everything published openly. The jump from minimum to ideal is mostly reach — more languages, more countries, more trained groups, and follow-through support so the capacity sticks.

Track record

The EduCampaigning method (Education → Workshops → Public Campaigns → Social Change) has been running for over fifteen years through the Latte Creative and Via Lattea team, across campaigns on AI safety, climate, and human rights. Clients include Greenpeace International, Save the Children, Doctors Without Borders (Medici Senza Frontiere), Oxfam, WWF, Council of Europe. Via Lattea APS was set up in 2025 as the campaigning continuation of that work. We're part of the Stop Killer Robots coalition (260+ organisations, 70+ countries).

Team

Salvatore Barbera runs it. Physics degree from Bologna, then twenty years in campaigning. Greenpeace across International, Mediterranean, and Italy offices from 2006 to 2012 (including the 2011 Italian nuclear referendum campaign). Country Director of Change.org Italy, launched it and grew it to 7M+ users. Co-founder of Riparte il Futuro, an Italian anti-corruption nonprofit that gathered over a million signatures. Founded Latte Creative in 2010 and Via Lattea in 2025. Salvatore attended the BlueDot AGI Strategy course this June, now on their Technical AI Safety and AI Governance courses. Sits on the doctoral board of the PhD in Design for Social Change at ISIA Roma.

Jenna Jauhiainen is senior campaign strategist and lead facilitator. She designs the co-design workshops that sit at the heart of the method. Her job is to make sure partner organisations walk out able to run their own campaign without us.

Luca Santarelli works on curriculum and research. PhD candidate at ISIA Roma, same programme as above.

Riccardo Botta handles messaging and the plain-language translation of AI safety concepts.

Giulia Zannoni does visuals and design. She's behind the "No machine can decide to kill" identity for Stop Killer Robots in Italy this September.

Failure modes

The main risk is uptake: we build the materials and nobody uses them. That's why we've scoped one pilot workshop at the minimum tier, to test it before scaling. If two partner organisations don't come back for more, we stop.

The other risk is that the loyalty bottleneck framing doesn't work outside AI safety circles. If so, the curriculum on autonomous weapons is still useful on its own.

Everything we produce is open source. Even in the worst case, the materials stay available.

Money raised

Via Lattea started in 2025. In the last months we've raised about €3,000 from individual Italian donors. The grantmaking.ai grant is our first institutional grant. We also have an application in with Longview Philanthropy for a bigger programme that would use this Manifund grant as its training core.

Links

Detailed budget: https://docs.google.com/spreadsheets/d/18wi34LPB4PWatnK5_9SYei-xDLKsSHaYMaCXc7EbCUs/edit?gid=0#gid=0

grantmaking.ai project: https://app.grantmaking.ai/projects/63b9965c-3e31-456d-b5cc-979aa6352a60