Systemic Altruism
← Back to the cohort
Justice & public goodsPreliminary editorial review

Partnership on AI

Partnership on AI convenes civil society, researchers and companies to develop guidance and shared practices for responsible AI. The hypothesis concerns common institutional capacity, not a regulator’s authority or automatic compliance.

Visit the organization ↗

Systemic Altruism practice potential: 70–80/100
90% of weighted axes reviewed. A mechanism hypothesis, not proven effectiveness or a donation recommendation.

Who this work enables

Tools & support for charities: Shared guidance, research and cross-sector working practices can help nonprofit advocates and public-interest implementers avoid developing AI governance methods alone.

Humanity’s Checklist source match

Coordination · Gatekeepers — editorial, partial alignment.

Convening across institutional boundaries and making governance guidance public can coordinate practice and contest information asymmetries. Corporate participation alone does not show power redistribution.

Explicit checklist adoption: not verified. A mechanism match is not an affiliation, endorsement, evidence of effectiveness or a scoring bonus. Read the practice.

Focus classification: 2026-10-27. Based on the organization’s dated official sources below; earlier axis-review dates are unchanged.

Compare alternatives for a funder →

The proposed pathway

Create shared AI governance guidance and cross-sector implementation capacity

Six dimensions, with rationale

Structural depth

3 / 4

Shared governance guidance aims to change organizational practices and incentives around AI, although voluntary recommendations are not binding institutional reform.

Institutional reach

4 / 4

Cross-sector convening can connect industry, civil society and research institutions; partner membership is not proof of implementation or a population-level effect.

Durability

2 / 4

Guidance can be reused in institutional practice, but voluntary commitments, fast-changing technology and ongoing coordination limit persistence without continued effort.

Scalability

3 / 4

Published resources and common practices can spread across organizations, while implementation capacity, conflicts and local regulation constrain straightforward replication.

Funding additionality

Unreviewed

Unknown: philanthropic and corporate contributions do not establish a marginal funding gap, unused programme capacity or what an extra donation would uniquely change.

Community agency

Unreviewed

Unknown: a balanced company/nonprofit board and stakeholder consultation do not establish binding decision rights for people directly affected by AI systems.

Evidence and selection

Primary-source programme and identity documentation; no independent impact validation in this review

Illustrative inclusion to examine this mechanism. Inclusion is not an external recommendation or a verified comparative ranking.

A testable next question

For organizations implementing a specific PAI guidance resource, do independently observed practices and affected-user outcomes improve relative to comparable non-adopters after accounting for selection and industry funding relationships?

Review confidence: Low–medium for the documented mechanism; low for causal effectiveness and donation comparisons. PAI’s current cross-sector guidance and convening capacity, excluding regulatory powers, every partner’s conduct and the separately assessed AI Incident Database operator.

What remains uncertain

Partnership on AI to Benefit People and Society is assessed as the US 501(c)(3) nonprofit described on its site. Corporate funding and equal for-profit/not-for-profit board representation are disclosed; independence claims are not an external conflict-of-interest audit. Voluntary guidance is not enforcement or demonstrated reduction in harms. The separately assessed Responsible AI Collaborative’s current AI Incident Database work is outside this scope; historical PAI involvement is not counted twice. Public guidance is not a named software license, and no code badge is assigned.

Sources to inspect

Propose a sourced correction →

Axis review: 2026-10-27 · weighting method v0.2 (2026-10-26) · Method, conflicts and coverage