The 6-Pack of Care

Most debates about “aligning” AI to human values try to solve it from the top down: write better rules, infer better preferences, train better models. Civic AI, as Audrey Tang and Caroline Green use the term, starts somewhere else. It asks who gets heard, who is accountable, how failures are repaired, and when a system should be switched off — and it answers by building many small AIs a community can own, inspect, correct, and shut down, rather than one powerful system built to govern everyone.

Each of these bounded local stewards has a name in their framework — a Kami — a spirit of a particular place, not a universal governor. The 6-Pack of Care is the test for whether one can be trusted.

The six packs

The first four form a loop — the care cycle — that keeps a system answerable to the people it affects. The last two make that loop hold across many organisations and over time.

1. Attentiveness — noticing what the people closest to a problem see that institutions miss.
2. Responsibility — who is accountable, with what authority, and what happens if they fail.
3. Competence — whether the system actually works in practice: audited, explainable, and safe to fail.
4. Responsiveness — whether affected people can contest an outcome and force a repair.
5. Solidarity — whether the wider ecosystem rewards cooperation, exit, and public accountability over lock-in.
6. Symbiosis — whether the system stays bounded, local, and sunset-ready instead of hardening into permanent rule.

Alignment as a public process

The load-bearing idea is that alignment is not solved once and installed. It is maintained in public — an accountable civic procedure the authors call alignment-by-process. This complements the safety work done inside frontier models; it does not replace it. Where a frontier lab asks “how do we build one aligned system,” the 6-Pack asks how to keep answerability in the rooms where deployed systems already act: care homes, classrooms, councils, deliberation platforms. It is the same instinct behind Audrey Tang’s plurality — spread capability sideways, steer it with communities — applied to the governance of AI itself.

The framework draws openly on Joan Tronto’s ethics of care, deliberative democracy, and Plurality. Its authors are blunt about the risk of empty relabelling:

The test that keeps it honest

If a principle does not change who can object, who must answer, what gets logged, or what happens after a failure, it is decoration.

Where it comes from

The 6-Pack grew out of real deployments on “unusually prepared soil” — places with civic hackers, responsive institutions, and strong community life. Its case studies include Taiwan’s Alignment Assemblies (which used Polis-style sensemaking to respond to a wave of AI-enabled scam ads), Team Mirai in Japan, Engaged California, the Tibetan-language Monlam AI, and the ROOST federated safety network. The authors are careful that each is proof of a mechanism, not a plug-and-play template: in a new setting you start again by asking who is missing, who has standing, and who can stop the system.

Civic AI in this sense is less about smarter chatbots than stronger self-government: institutions that show their work, repair harm in public, and carry civic memory across generations. It is a close cousin of conversation networks and the wider argument that AI is a social technology to be governed, not an autonomous mind to be contained.

Sources

• Civic AI: the 6-Pack of Care — Audrey Tang & Caroline Emmer De Albuquerque Green, Oxford Institute for Ethics in AI. Framework site with the manifesto and per-pack pages; the book follows from Wiley in February 2027. Presented at the Civic AI Conference, Rhodes House, Oxford, 25 March 2026.
• “AI and the Future of Digital Public Squares” — Collective Intelligence (2026): a peer-reviewed agenda, co-authored by Tang with Beth Goldberg and Hélène Landemore among others, that cites the 6-Pack of Care directly.