
When people want different things, the reflex is to treat it as a war to win or a pie to split. But most stand-offs are only zero-sum on the surface: look behind each position to the need beneath it, and a larger option can often satisfy everyone. Solidarity is the discipline of finding those win-wins and, more durably, of building an ecosystem where the cooperative move is also the easiest one.¶ 1 (permalink to paragraph 1)
Traffic is safer because roads, signs and rules make safe driving easier than reckless driving. The drivers do not have to be saints.¶ 2 (permalink to paragraph 2)
Even perfect local care fails in a hostile wider ecosystem. Solidarity equips the field so that civic behaviour wins by design.¶ 3 (permalink to paragraph 3)
Start in your room
Ask, “Can we leave without losing the people and records we depend on?” Before making a system mandatory, have the people who use it try an export and a handover to another service. Cooperation needs agreements across groups as well as shared software.¶ 4 (permalink to paragraph 4)
In Caring Democracy (p. 23), Joan Tronto’s fifth phase, caring with, asks whether care is consistent with democratic commitments to justice, equality, and freedom for all. She takes the phase’s moral qualities, plurality, communication, trust, and respect, from Selma Sevenhuijsen’s Citizenship and the Ethics of Care and adds solidarity (p. 35). For Civic AI, solidarity asks a structural question: do the rules of the ecosystem make cooperation easier to practise than domination?¶ 5 (permalink to paragraph 5)
There is a technical reading of that question. In the sequential social-dilemma simulations studied by Leibo and colleagues in 2017, independently rewarded agents could cooperate or defect depending on the game and their capabilities. Trivedi and colleagues’ Solipsistic Superintelligence is Unlikely to be Cooperative examines why unilateral optimisation can fail among interdependent actors. Taniguchi and colleagues propose a different objective: a collective term that no single agent can satisfy alone. This is one theoretical account of solidarity. It is no evidence that separately rewarded systems must polarise, or that this term is needed for civic cooperation.¶ 6 (permalink to paragraph 6)
Definition
Identity without exposure. Agent IDs should prove that an agent answers to someone accountable without publicly exposing private details. This is selective disclosure, or meronymity, the term Soliman and colleagues coined at CHI 2024 for revealing chosen facets of identity with trusted endorsers: proving role without revealing person. Accountability should not require doxxing.
¶ 7 (permalink to paragraph 7)Interoperability beats captivity. Portability and open protocols move competition to quality of care. Exit rights protect freedom by making it possible to leave without losing your audience or history: Albert Hirschman’s exit, which disciplines an incumbent only when leaving is practical.
Utah’s Digital Choice Act (H.B. 418, 2025; now ch. 13-81) already requires a portable copy of a user’s personal data, including the social graph (§13-81-201, in force 1 July 2026), to the extent technically feasible and readily usable to the extent practicable. Interoperability in the current text is synchronous sharing (§13-81-202), with carve-outs for inferences and ranking systems. H.B. 408 (2026) replaces that from 1 July 2027 with continuous real-time forwarding.
¶ 8 (permalink to paragraph 8)Federation over monoliths. Share threat intelligence without a single chokepoint. Local policies stay local, but defence compounds across institutions.
¶ 9 (permalink to paragraph 9)Expression is not amplification. Recommender accountability is a civic duty. Free speech is not free reach, as Renée DiResta put it in 2018, and the argument of Sunset Section 230. Ranking in civic contexts should reward cross-group reason-giving and shared problem-solving, not only whatever inflames one cluster fastest. California’s SB 1159, signed into law in September 2026, excludes AI systems from public-member standing while allowing people to use assistive tools at ordinary human volume. Standing to speak in civic channels belongs to humans who can prove it without surrendering identity; algorithmic multiplication earns no such standing.
¶ 10 (permalink to paragraph 10)
Why it matters
Local repair is fragile if the wider infrastructure rewards lock-in, opacity, and outrage. A system can be attentive, responsible, competent, and responsive inside one deployment while still being trapped in a hostile market or protocol environment.¶ 11 (permalink to paragraph 11)
Solidarity means no one gets a structural exemption from caring. Tronto’s metaphor: “The first thing that we need to do is to collect all of those free passes out of taking care responsibilities seriously.” For AI systems, Tronto’s passes come back in new words: “we’re too big to regulate,” “users consented in the ToS” (the bootstrap pass), “we opened the weights,” and “we donated to safety research” (the charity pass). The passes AI adds of its own are named in Pack 2. Solidarity collects these passes by making cooperation cheaper than defection, without pretending the architecture makes capture impossible.¶ 12 (permalink to paragraph 12)
Solidarity is where the framework becomes democratic infrastructure. It asks whether justice, equality, freedom, plurality, and mutual accountability are easier to practise because of the ecosystem, or in spite of it.¶ 13 (permalink to paragraph 13)
Open stacks and anti-rival resources
Carnegie funded more than 1,600 public libraries in the United States. What made them democratising was the open stacks, where patrons browsed the shelves themselves and did not have to petition a librarian to fetch on their behalf. Vertical AI platforms that hold capability behind proprietary APIs reproduce the closed-stack model; open-weight models, open protocols, and shared eval registries are the open-stack equivalent: infrastructure that treats communities as capable actors who set their own terms.¶ 14 (permalink to paragraph 14)
Knowledge, unlike oil, is anti-rival: Steven Weber’s term from The Success of Open Source, meaning that use enriches rather than depletes it. The same holds for open protocols: eduroam grows more valuable for every member as institutions join, with no participant capturing the gain. A solidarity-equipped ecosystem works the same way: when one city adopts selective-disclosure attestations, every other city can adopt the standard at near-zero marginal cost.¶ 15 (permalink to paragraph 15)
None of this emerges spontaneously. It takes law as infrastructure: interoperability-as-public-good legislation, of which Utah’s Digital Choice Act is a live portability statute, not yet the 2027 continuous-forwarding duty. Solidarity is where the individual care cycle connects to the commons. It draws on Elinor Ostrom’s design principles for durable commons and her later account of polycentric governance: many centres of decision-making, overlapping, none sovereign over the rest.¶ 16 (permalink to paragraph 16)
What it looks like in practice
Selective-disclosure identity. AI systems, organisations, and people have verifiable attestations held by trusted custodians; public proofs stay minimal, but challenge and revocation remain real.
¶ 17 (permalink to paragraph 17)Social portability. Users export social graph and content, pass interoperability tests, and keep personal audiences when leaving. Freedom becomes practical because exit no longer means social exile. Vendor lock-in turns a commercial inconvenience into a care relationship people cannot withdraw from.
¶ 18 (permalink to paragraph 18)Uncommon-ground audits. Platforms publish an uncommon-ground index, the share of shared decisions that earn real cross-group participation and endorsement. It is the same cross-group signal that bridging-based ranking reads to sort feeds, raised here from ranking policy to public audit. Plurality becomes visible instead of resting on branding claims. The index does not replace genuine pluralism, but it makes pluralism harder to fake. Read it against Pack 1’s representation gap, because the cheapest way to raise co-endorsement is to narrow who counts.
¶ 19 (permalink to paragraph 19)Federated safety network. Partners detect harms in their own cultural context, share what they can through open tools, and keep enforcement local. No single hub holds everything. ROOST shows the reusable, self-hostable layer of that aim (Osprey for incident response, Coop for review), and its Model Community shares open-weight safety models and policy packs, the distribution and practice layer that makes federated training of safety models feasible for platforms that cannot pool their raw data.
¶ 20 (permalink to paragraph 20)Protocol-level norms. Machine-readable terms of cooperation set common duties: no scraping without consent, honour appeal handoffs, respect exit.
¶ 21 (permalink to paragraph 21)
We propose that coalitions of public buyers and intermediary institutions fund and require this common layer of permissions, provenance, redress and exit. No single lab should own it; this is a proposal for shared infrastructure, not an implemented system.¶ 22 (permalink to paragraph 22)
From ideas to practice
- Stand up an identity custodian. Set up community orgs or public interest entities to issue and hold attestations; publish revocation and challenge endpoints.¶ 23 (permalink to paragraph 23)
- Mandate portability in procurement. Public buyers require protocol interoperability, handover rehearsals and exit drills as standard contract terms.¶ 24 (permalink to paragraph 24)
- Adopt uncommon-ground audits. Ensure platforms publish the uncommon-ground index and its supporting diagnostics quarterly; have third parties verify.¶ 25 (permalink to paragraph 25)
- Join a safety federation. Contribute to and consume from a shared threat registry; localise enforcement.¶ 26 (permalink to paragraph 26)
- Default to civic ranking rules. In civic contexts, make feeds reward reason-giving, contestability, and cross-group cooperation, and stop rewarding outrage.¶ 27 (permalink to paragraph 27)
Buildable tools
- Agent identity schema for selective disclosure, revocation lists, and proof formats.¶ 28 (permalink to paragraph 28)
- Portability harness for export/import scripts and fidelity checks.¶ 29 (permalink to paragraph 29)
- Uncommon-ground audit kit to compute and publish cross-group participation and endorsement metrics.¶ 30 (permalink to paragraph 30)
- Federated safety hub with open APIs for threat intelligence and local adapters.¶ 31 (permalink to paragraph 31)
- Shared cooperation rules that AI systems and platforms can verify automatically.¶ 32 (permalink to paragraph 32)
One case: the flood-bot
This fictional composite is a teaching illustration. It records no actual deployment or measured results.¶ 33 (permalink to paragraph 33)
The crisis deepens, and the community has now renamed the bot River-Steward: an attentive, limited-scope AI system stops looking like a product and starts looking like a neighbour.¶ 34 (permalink to paragraph 34)
- Identity. Scammers begin impersonating aid workers to intercept funds, while real volunteer translators face harassment from stressed, displaced claimants. The city shifts to selective-disclosure IDs so translators can prove their role without exposing personal phone numbers. Abuse drops; accountability rises.¶ 35 (permalink to paragraph 35)
- Portability. Procurement rules require case files, consent records, and contacts to move from the emergency bot to long-term housing services in one export rather than being typed in again.¶ 36 (permalink to paragraph 36)
- Federated defence. Scam reports from neighbouring cities flow in via the safety network; the bot downgrades suspect links by default.¶ 37 (permalink to paragraph 37)
- Uncommon-ground audit. A weekly public uncommon-ground index shows whether renters and homeowners are co-endorsing more shared remedies than at launch.¶ 38 (permalink to paragraph 38)
What could go wrong
- ID creep. IDs become dossiers. Fix: Selective disclosure by design; minimal proofs; independent custodians; strong revocation.¶ 39 (permalink to paragraph 39)
- Portability theatre. Exports are unreadable or lossy. Fix: Exit tests in contracts; penalties for fidelity failure.¶ 40 (permalink to paragraph 40)
- Federation capture. One big player dictates norms. Fix: Polycentric governance, in Ostrom’s sense; open standards; rotating custodians.¶ 41 (permalink to paragraph 41)
- Federation as attack surface. Without security hardening, a network of local AI systems becomes a distributed botnet waiting to be recruited. Fix: Mandate sandboxing, least-privilege execution, and input validation at every node (Pack 3); federated security audits; shared vulnerability disclosure.¶ 42 (permalink to paragraph 42)
Interfaces
- From Attentiveness (Pack 1): the bridging map hands over the terrain: where groups agree, where they clash, and where a crossing is possible. The uncommon-ground index counts the crossings actually made; composition precedes co-action.¶ 43 (permalink to paragraph 43)
- From Responsibility (Pack 2): portability and exit clauses are referenced in every engagement contract.¶ 44 (permalink to paragraph 44)
- From Competence (Pack 3): dependable, well-instrumented systems are what make public audits credible.¶ 45 (permalink to paragraph 45)
- From Responsiveness (Pack 4): repair culture feeds trust across organisations.¶ 46 (permalink to paragraph 46)
- To Symbiosis (Pack 6): solidarity provides the treaties local systems need to cooperate.¶ 47 (permalink to paragraph 47)
A closing image: the quilt without a backing
Imagine a quilt that needs no single backing cloth to hold together. Each square is sewn by a different hand, for a different room, and the squares join edge to edge. Pull on one square and the rest come with it—not because someone commands them to but because they are joined. A federated safety network is a quilt: many local stewards, many local norms, one shared seam. The quilt with no backing is the image we hold for solidarity.¶ 47a (permalink to paragraph 47a)
Sources for this pack
- Tronto, Caring Democracy (2013), pp. 23, 35, and 169: caring with, its moral qualities, and collecting the passes; Sevenhuijsen, Citizenship and the Ethics of Care (1998).¶ 49 (permalink to paragraph 49)
- Hirschman, Exit, Voice, and Loyalty (1970): exit as discipline.¶ 50 (permalink to paragraph 50)
- Weber, The Success of Open Source (2004): anti-rival goods; Benkler, The Wealth of Networks (2006): open stacks.¶ 51 (permalink to paragraph 51)
- Soliman et al., “Meronymous Communication” (CHI 2024): meronymity.¶ 52 (permalink to paragraph 52)
- DiResta, “Free Speech Is Not the Same As Free Reach” (2018); Lanier, Stanger, and Tang, “Sunset Section 230” (2026): expression is not amplification.¶ 53 (permalink to paragraph 53)
- Ovadya, Bridging-Based Ranking (2022): civic ranking rules.¶ 54 (permalink to paragraph 54)
- Ostrom, Governing the Commons (1990) and “Beyond Markets and States” (2010): commons design principles and polycentric governance.¶ 55 (permalink to paragraph 55)
- Leibo et al., “Multi-agent Reinforcement Learning in Sequential Social Dilemmas” (2017); Taniguchi et al., “Symbiotic Alignment via Collective Predictive Coding: A Theoretical Framework for Co-Creative Human–AI Ecosystems” (2026); Trivedi et al., “Solipsistic Superintelligence” (2026): a theoretical technical reading, not a field result.¶ 56 (permalink to paragraph 56)
- Utah H.B. 418, the Digital Choice Act (2025), now §§13-81-201 and 13-81-202 (in force 1 July 2026; superseded 1 July 2027); H.B. 408 (2026): portability now; continuous real-time forwarding from 1 July 2027.¶ 57 (permalink to paragraph 57)
- The full trail is on the Sources page.¶ 58 (permalink to paragraph 58)
