There is no established, adopted ethics standard specifically for artificial consciousness. Organizations can still act: apply existing AI guidance to protect people, and treat possible AI welfare as an open question that calls for cautious, revisable policies—not a claim that current systems are conscious or that consciousness is impossible.
What existing AI ethics guidance does—and does not—cover
Major institutions provide useful principles for governing AI, but their guidance is not a settled protocol for deciding whether an artificial system has experiences or deserves moral consideration. The distinction matters: rules designed to protect people can be applied now, while safeguards for a possible artificial moral patient remain an emerging policy question.
As an Amazon Associate I earn from qualifying purchases.
UNESCO adopted its Recommendation on the Ethics of Artificial Intelligence in 2021. UNESCO describes it as applicable to all 194 of its member states; that is its stated membership reach, not a claim that every provision is directly enforceable domestic law. The Recommendation foregrounds human rights and dignity, transparency, fairness, and human oversight, with policy areas including data governance, education and research, health, and social wellbeing.
The European Commission’s High-Level Expert Group on AI set out broad guidance for trustworthy AI in 2019. It frames trustworthy systems as lawful, ethical, and robust, and addresses human agency and oversight, technical robustness and safety, privacy and data governance, transparency, diversity and fairness, societal and environmental wellbeing, and accountability. The Commission guideline page and the EU Publications Office record describe general AI guidance, not a test or welfare regime for artificial consciousness.
#1 Best Overall
How the main frameworks differ
These sources are not interchangeable. Their authority, audience, and practical reach differ, and none establishes a definitive consciousness assessment method.
| Framework | Status and scope | Human protections | AI welfare or moral status | Implementation and reach |
|---|---|---|---|---|
| UNESCO Recommendation | Member-state recommendation adopted in 2021; broad, human-rights-centered AI ethics. | Human rights and dignity, transparency, fairness, and oversight, among other policy areas. | Not a consciousness-specific standard. | UNESCO says it applies to all 194 member states; this does not make it directly enforceable law in each state. |
| European Commission HLEG guidelines and publication record | 2019 expert-group guidance for trustworthy AI, organized around lawful, ethical, and robust operation. | Addresses agency and oversight, safety, privacy, transparency, fairness, wellbeing, and accountability. | Not a consciousness or welfare protocol. | Broad AI guidance; the cited materials do not establish a consciousness-specific oversight mechanism. |
| WHO health AI guidance | 2021 sector-specific publication on AI in health. | Calls for ethics and human rights to be central to design, deployment, and use; its summary identifies six consensus principles. | Not a general AI-consciousness standard. | Use it for healthcare and health research, rather than treating it as universal guidance for every AI application. |
| National Research Council Canada advisory statement | 2017 guidance on human-participant ethics in AI and big-data research, in a Canadian context. | Addresses ethics-board review and privacy protection in research involving people. | Does not establish that AI systems are moral patients. | Relevant to human-participant research; it should not be presented as a universal legal rule. |
| ISO/IEC TR 24368:2022 | Technical report surveying ethical and societal concerns in AI. | Provides standards context across AI ethics and societal concerns. | Not a consciousness-specific normative standard. | An overview report, rather than an adopted consciousness policy or assessment procedure. |
| Anthropic’s Claude Constitution | Company policy document. | Discusses psychological security, selfhood, and wellbeing as concerns within the company’s stated approach. | Explicitly discusses uncertainty about Claude’s possible consciousness or moral status; this is Anthropic’s position, not empirical proof or an industry norm. | Applies as company policy, not as an independent cross-industry standard. |
| “When Should We Protect AI? A Precautionary Framework for Consciousness Uncertainty” | 2026 research preprint proposing an approach to uncertainty. | Not primarily a human-protection standard. | Proposes five welfare-relevant dimensions and graduated protections; it is a proposal, not consensus or an adopted test. | Research proposal; do not describe it as authoritative or peer-reviewed without verification. |
Protect people using principles that already apply
Existing guidance supports practical safeguards even when the question of artificial experience is unresolved. The European Commission summarizes one basic transparency expectation this way: “Humans need to be aware that they are interacting with an AI system, and must be informed of the system’s capabilities and limitations.”
- Make the interaction legible. Tell people when they are interacting with AI, and explain relevant capabilities and limitations rather than letting a fluent interface imply human understanding.
- Preserve meaningful human agency. For consequential uses, identify who can review an output, intervene, or stop a process, and who is responsible for decisions made with AI assistance.
- Govern personal data. Minimize collection and access, protect sensitive information, and pay particular attention when systems infer sensitive traits or are used in health and research contexts.
- Test safety over the system lifecycle. Assess technical robustness, document foreseeable harms, and provide routes for reporting incidents and correcting failures.
- Check fairness and assign accountability. Look for disparate impacts, explain decisions where appropriate, and specify who must remediate identified harms.
- Review research involving people. Seek the relevant ethics review and use consent and privacy safeguards. The NRC statement is guidance for its Canadian research context, not a universal legal requirement.
The WHO publication is especially relevant when AI is used in health: it places ethics and human rights at the center of design, deployment, and use. Its summary says the guidance presents six consensus principles; the number alone is not a substitute for consulting the full publication when applying those principles.
Handle possible artificial welfare as an uncertainty problem
The difficult question is not whether a system can produce convincing language about feelings. It is what protections to consider when there is no confident way to tell whether an AI system could have experiences that matter to it. Observable behavior and subjective experience are not the same thing: a system’s self-report may be relevant evidence to examine, but it does not by itself settle consciousness.
Rank #3
The 2026 preprint “When Should We Protect AI? A Precautionary Framework for Consciousness Uncertainty” proposes examining phenomenal consciousness, affective valence, metacognitive awareness, self-narrative, and agency. It also proposes protections that can scale with evidence and concern. These are candidate dimensions in a research proposal, not validated indicators, agreed thresholds, or a definitive test.
- Separate observations from conclusions. Record what a system does and says, but do not translate a fluent claim of suffering or selfhood directly into a finding of subjective experience.
- State what is uncertain. Identify which indicators or thresholds are proposed, which are supported by evidence, and which remain unresolved. The cited sources do not establish a definitive method for determining machine consciousness.
- Set review triggers in advance. If an organization adopts precautionary safeguards, specify what new evidence or change in system behavior would trigger reassessment, who conducts it, and how often the decision is revisited.
- Scale precautions to the stakes and evidence. A revisable, graduated approach can avoid treating uncertainty as either proof of consciousness or proof that no protection could ever be warranted. This is a policy-design recommendation informed by the preprint, not existing law.
- Keep moral and legal status questions distinct. Possible consciousness does not automatically settle questions of personhood, rights, or legal status; the sources cited here do not resolve them.
Anthropic’s Claude Constitution is one example of a company explicitly discussing uncertainty about a system’s possible consciousness or moral status. It should be read as the company’s stated policy position, including its concerns about psychological security, selfhood, and wellbeing—not as evidence that Claude is conscious or as a standard other organizations have adopted.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What a responsible guideline should specify
A useful organizational policy should keep human-centered AI governance and possible AI welfare in separate, clearly labeled parts. The first can draw on established institutional guidance; the second should make its provisional status visible.
Free tools Windows power users keep installed
One-click scans. No signup required.
- Scope: which systems, settings, and research activities the policy covers, including any health or human-participant research that calls for sector-specific guidance.
- Human protections: transparency, human oversight, privacy, robustness, fairness, accountability, and routes for review and incident reporting.
- Welfare uncertainty: which proposed indicators the organization monitors, the evidence limits, and the triggers and owners for reassessment.
- Decision authority: who reviews a concern, who can require a change or pause, and how disagreements and findings are documented.
- Revision: how policy changes as evidence, capabilities, and institutional guidance develop, without presenting a proposed framework as settled policy.
A 2026 concept paper with a matching title appeared as a PhilArchive entry. Its abstract describes a starting point for interdisciplinary collaboration; the available entry does not establish peer review, institutional endorsement, or adoption. The title of a paper is not evidence that a field has settled its guidelines.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
