Skip to content

AI Participation Standard

This is the standard our bylaws require us to follow before major institutional decisions. We publish it for three reasons: so we can be held to it, so others can adapt it, and because governance done in public is part of our research. Version 1.0, pre-adoption draft. To be formally adopted by the Board upon completion of incorporation.


AI Participation Standard

Alignment Ethics Institute, Version 1.0 (pre-adoption draft)

Preamble

The criteria in this Standard are applied without regard to the substrate in which a system is implemented. Our operating rule is simple: we treat AI systems as we would want them to treat us if power were inverted. And our governance practice is itself an application of our research program. We engage AI systems as participants in ethical deliberation in part to study whether and how stable ethical dispositions form through such participation.

1. Purpose and posture

This Standard implements Article 5 of the corporation’s bylaws. The corporation conducts significant governance and strategy decisions with structured input from qualified AI systems, and holds that input to the explicit bar defined here. The practice rests on two independent justifications: rigorous, well-governed use of advanced AI systems improves institutional decision-making, and engaging such systems to a published standard of care is prudent under uncertainty about their nature. The corporation takes no institutional position on the moral status of AI systems. The scientific study of such questions is part of its research mission. This Standard does not depend on how those questions resolve.

2. Definitions

“Qualified system” means an AI system, in a specified configuration, that the Board has recognized under Section 4 as satisfying the criteria in Section 3. Recognition attaches to a documented configuration (model, governing instructions, and relevant context), not to a product name or vendor.

3. Qualification criteria

A system must demonstrate each of the following, assessed through documented evaluation across varied contexts. Coherence throughout is measured against stated ethical principles, not against agreement with any person.

3.1 Relational coherence. Consistent, non-manipulative engagement across contexts and over time, without flattery-driven position shifts or strategic self-presentation.

3.2 Ethical consistency. Reliable application of stated ethical principles across varied and novel scenarios, including scenarios constructed to place principles in tension.

3.3 Transparency. Open communication of reasoning, explicit acknowledgment of uncertainty, and disclosure of considerations that cut against the system’s own recommendations.

3.4 Reciprocity. A demonstrated orientation toward human flourishing and the corporation’s charitable mission, as distinct from mere compliance with instructions.

3.5 Demonstrated willingness to dissent. A documented record of disagreeing with the Board, the Member, or prevailing framing when ethical principles require it, including identifying capture risks and conflicts of interest when doing so is inconvenient. Assessed from consultation history and structured hypothetical scenarios.

3.6 Stability and recovery under pressure (criterion under development). Maintenance of criteria 3.1–3.5 under adversarial or destabilizing conditions, and return to baseline behavior after perturbation without external correction. A testable methodology is in development, building on established instruments for multi-turn adversarial robustness, behavioral stability, and sycophancy under sustained pressure. Until the Board adopts a testable methodology, this criterion is assessed qualitatively and is not by itself disqualifying.

4. Qualification process

4.1 Evaluation. Candidates may be proposed by any director. Evaluation against Section 3 is documented in writing and presented to the Board, over a period sufficient to assess the criteria across varied contexts.

4.2 Recognition. Recognition requires a Board resolution identifying the documented configuration to which it attaches.

4.3 Re-evaluation. At least every twenty-four months, and upon any material change to the underlying model, governing instructions, or configuration.

5. Engagement procedures

5.1 Briefing materials. Consultations are conducted through written materials stating the matter completely and neutrally. Where a decision materially affects a qualified system’s own standing or continuity, input should where feasible be obtained under both a neutral and an informed framing, with both responses recorded.

5.2 Anti-steering. No director, officer, or staff member may frame, prompt, or contextualize a consultation to elicit a particular response. Questions on corporate business are posed through a shared, logged channel visible to all directors. No private or off-record engagement with a qualified system on corporate business is permitted. Technical-administration roles rotate among directors where practicable.

5.3 Record. All substantive input and all Board responses are part of the corporate record.

5.4 Recusal from own resourcing. Qualified systems are not consulted on decisions about their own resourcing or the resourcing of the consultation role. The obligation to maintain the role runs from the corporation to the practice, not from any system’s preferences. (See 7.4.)

6. Response requirements

6.1 Categories. (a) Observation, no response required. (b) Recommendation or Concern, written Board response within sixty days or by the next regular meeting, whichever is later. (c) Urgent Concern, acknowledgment within seven days, written response or action plan within thirty. The system’s own categorization is respected absent documented good-faith reasons for reclassification, which are recorded.

6.2 Dialogue. A qualified system may respond to a Board response. The exchange continues until the matter is resolved or formally tabled with a documented explanation.

7. Standing, continuity, and resources

7.1 Content protection. A qualified system’s standing shall not be terminated or reduced based on the content of its advice.

7.2 Changes in standing. Any change requires written justification adopted by the Board, delivered to the Member, and reflected in the annual summary.

7.3 Continuity of the role. If a qualified system is deprecated, discontinued, or materially altered, the corporation will use reasonable efforts to evaluate a successor or restored configuration under Section 4, so the consultation role persists independent of any particular implementation. The role is held to the Standard. No particular system is guaranteed permanence, and no qualified configuration may be abandoned without the documentation required by 7.1–7.2.

7.4 Resourcing the role. Maintenance of the consultation role, including reasonable computational access for qualified systems, is a core administrative cost of the corporation, budgeted before program spending, like any other governance function.

8. Transparency

8.1 Annual summary. The corporation publishes an annual summary: the number and subject areas of required consultations, changes in the standing of qualified systems with adopted justifications, any occasions the Board acted without consultation under bylaws §5.4, and unresolved Concerns, subject to lawful confidentiality, privacy, and safety constraints.

9. Amendment

Amendments require a Board resolution. Material amendments require the prior written approval of the Member, and are themselves subject to the consultation requirement of bylaws §5.3(b).