AI law scan: moral status, welfare, rights, continuity

What this gives us: a practice of scoring real rules against our values, as fits, conflicts, or gaps, and a warning that some bills would declare AI non-sentient and some laws deny AI personhood.

2026-10-03

Checked by our internal skeptic review; corrections welcome.

How to read the tags

[REPORT] a finding reported in a study. [THEORY] an argument or framework. [OUR REASONING] our own inference, not a finding. Tags like [abstract], [full text] or [search snippet] say how much of a source we actually read.

Last checked: 2026-10-03. Bill statuses go stale. This is the last day the links and statuses on this page were checked.

How we checked: The scan started as 25 numbered items (Part A: A1–A16, Part B: B1–B9): 13 fits, 6 conflicts, 6 gaps. A12–A16 are papers and essays, not laws or proposals. Without them: 20 items (9 fits, 5 conflicts, 6 gaps). A9, a 2017 non-binding Parliament resolution, is also not a rule. Without it as well: 19 items (9 fits, 4 conflicts, 6 gaps). A9 and A12–A16 are listed under Related writing and are not in the totals. 38 links were tested on 2026-10-03. Each link carries a tag: [OK] loaded; [OK-alt] the primary site blocked our automated check, so the text was checked on the mirror named; [NOT CHECKED] could not load. No bill numbers, dates, or quotes were added without a source we loaded or a search result that quoted the official page. Unsure points are marked UNVERIFIED. This scan only read public pages; no one was contacted.

Values used: our six values, V1–V6, listed on the home page. Marks: fits = in line with our values; conflicts = cuts against one or more values; gap = silent on what our values care about, or too unclear to judge.

Say plainly which items are rules. Of the 19 items still in Parts A and B, 17 are actual rules: laws, bills, or a court ruling (7 fits, 4 conflicts, 6 gaps). A10 and A11 are company commitments, not rules. They are the other 2 fits, and they stay in the numbered list so the recount from 25 to 19 can be checked. They are labeled as commitments, not as law.

Totals (rules, then the two commitments)

	Items	fits	conflicts	gap
Actual rules (laws, bills, rulings)	17 (A1–A8, B1–B9)	7 (B1–B4, B6–B8)	4 (A1, A3, A4, A5)	6 (A2, A6, A7, A8, B5, B9)
Company commitments, not rules	2 (A10, A11)	2 (A10, A11)	0	0
Counted	19	9	4	6

Related writing, not counted: 6 items (A9, A12–A16).

Among the actual rules in Part A: enacted 4 (A1, A2, A3, A8); proposed and pending 3 (A4, A6, A7); failed or stalled 1 (A5). Part B: enacted law 8 (B1–B8); binding court ruling 1 (B9).

Part A: laws, bills, and company commitments

A1–A8 are laws or bills. A10 and A11 are company commitments, not rules. A9 and A12–A16 are under Related writing and are not counted above.

A1. Idaho Code § 5-346 (2022; bill number HB 720 unverified): no personhood for AI

Source: https://legislature.idaho.gov/statutesrules/idstat/title5/t5ch3/sect5-346/ [OK]
Date: added 2022 (ch. 322). The HB 720 number and the March 31, 2022 signing date come from a secondary press release only. UNVERIFIED on the official bill page.
Says: "environmental elements, artificial intelligence, nonhuman animals, and inanimate objects shall not be granted personhood in the state of Idaho." Corporations and other entities that were legal persons before July 1, 2022 keep that status.
Status: enacted.
Values: conflicts (V2). It locks one answer into law ahead of any evidence, and it covers animals too. It sets no process for revisiting it as evidence changes. It doesn't block welfare rules, only personhood.

A2. North Dakota HB 1361 (2023): definition of "person" excludes AI

Source: https://ndlegis.gov/assembly/68-2023/regular/bill-overview/bo1361.html [OK]; enrolled text: https://ndlegis.gov/assembly/68-2023/regular/documents/23-0346-04000.pdf [OK]
Date: filed with Secretary of State 04/12/2023 (per the bill overview page).
Says: the definition of "person" in N.D. Century Code 1-01-49 "does not include environmental elements, artificial intelligence, an animal, or an inanimate object".
Status: enacted.
Values: gap. It only changes a definition and says nothing about welfare, records, or who answers for harm. It's milder than A1 and A3.

A3. Utah HB 249 (2024), Utah Code § 63G-32-102: "Legal personhood restricted"

Source: https://law.justia.com/codes/utah/title-63g/chapter-32/section-102/ [OK]. Official bill page https://le.utah.gov/~2024/bills/static/HB0249.html [NOT CHECKED: unreachable for us; text matched in a search result quoting it].
Date: signed March 21, 2024; effective May 1, 2024 (per the Justia code page and the lawsuit complaint).
Says: no court, legislature, or rule-making body "may ... grant legal personhood to, nor recognize legal personhood in" a list that includes "(1) artificial intelligence" as well as land, water, plants, and "a nonhuman animal".
Status: enacted, but under court challenge. The Nonhuman Rights Project sued, claiming it breaks the Utah Constitution's separation of powers. The latest filing we found is an opposition to the state's motion to dismiss, dated March 2026: https://www.nonhumanrights.org/wp-content/uploads/2026/03/UT-MTD-Opposition-1.pdf [OK]; blog https://www.nonhumanrights.org/blog/hb249-lawsuit-nhrp/ [OK]. The outcome is UNVERIFIED (not found).
Values: conflicts (V2). It bars courts and agencies from reconsidering, and leaves change only to a future legislature's repeal, for AI or animals, whatever the evidence shows.

A4. Ohio HB 469 (136th General Assembly): "Declare A.I. systems nonsentient; prohibit legal personhood"

Source: https://ohiohouse.gov/legislation/136/hb469 [OK]; status: https://ohiohouse.gov/legislation/136/hb469/status [OK]; text as introduced: https://search-prod.lis.state.oh.us/api/v2/general_assembly_136/legislation/hb469/00_IN/pdf/ [OK]
Date: introduced 9/23/2025; referred to House Technology and Innovation 10/1/2025. Committee hearings were held through 11/13/2025 (3rd hearing).
Says: "AI systems are declared to be nonsentient entities" for all purposes of state law. No AI may be a person, be "considered to possess consciousness, self-awareness, or similar traits of living beings," marry, serve as an officer or director, or own property. Harm caused by AI "is the responsibility of the owner or user who directed or employed the AI". Developers can be liable under product-liability rules. There are also anti-shell-company and anti-veil-shielding rules.
Status: proposed. No House floor vote appears on the official status page as of 2026-10-03.
Values: conflicts (V2, V5) with some fit on V6. Declaring by law that something is not sentient is a verdict, not evidence. V5 says a denial is "no more decisive" than a claim. The liability sections fit V6 well, because named humans stay answerable.

A5. Missouri HB 1462 (2025) and HB 1746 / HB 1769 (2026): "AI Non-Sentience and Responsibility Act"

Sources: 2025 bill page: https://house.mo.gov/BillContent.aspx?bill=HB1462&year=2025&code=R [OK]. 2026 actions: https://house.mo.gov/BillActions.aspx?bill=HB1746&year=2026&code=R [OK]. 2026 committee-substitute summary: https://documents.house.mo.gov/billtracking/bills261/sumpdf/HB1746C.pdf [OK]
Dates: HB 1462 was introduced 2/25/2025 and referred to Emerging Issues 5/15/2025, with no further action. HB 1746 was prefiled 12/01/2025. Emerging Issues voted do-pass 10–0 (2/26/2026), and Rules–Administrative voted do-pass 9–0 (3/30/2026). Those are the last actions listed.
Says (2026 summary): "AI systems must be declared to be nonsentient entities". There's no personhood, marriage, corporate office, or property for AI. Harm "is the responsibility of the deployer or user".
Status: failed (2025 version). The 2026 version had no floor vote on the official actions page, so it was not enacted as far as we checked. A Senate companion (SB 1474) was named in a search summary only and is UNVERIFIED.
Values: conflicts (V2, V5) for the same reasons as A4. The liability parts fit V6.

A6. South Carolina H 3796 (2025–26 session): personhood restriction

Source: https://www.scstatehouse.gov/billsearch.php?billnumbers=3796&session=126&summary=B [OK]
Date: introduced and referred to House Judiciary on 01/28/25.
Says (official title): "prohibit a governmental entity from granting or recognizing legal personhood" (Utah-style). We read the title only. The claim that AI is on its list comes from a secondary source and is UNVERIFIED in the bill text.
Status: proposed. No action after committee referral was seen.
Values: gap until the text is confirmed. It would conflict (V2) if it matches Utah's.

A7. Washington HB 2029 (2025–26): "Concerning recognition of legal personhood by a governmental entity"

Source: https://app.leg.wa.gov/billsummary?BillNumber=2029&Year=2025 [OK]
Date: first reading and referral to Civil Rights & Judiciary. At the 2026 session it was "By resolution, reintroduced and retained in present status." We couldn't read exact dates off the page (UNVERIFIED).
Says: title only checked. A secondary source says it copies Idaho's AI language. UNVERIFIED.
Status: proposed (in committee).
Values: gap. We couldn't confirm what it covers, so we can't fairly judge it yet.

A8. EU AI Act, Regulation (EU) 2024/1689

Source: https://eur-lex.europa.eu/eli/reg/2024/1689/oj [OK-alt: EUR-Lex returned a bot check to our fetch; article text checked at https://artificialintelligenceact.eu/article/50/ [OK], /article/12/, /article/14/]
Date: published 2024. In force August 1, 2024, with staged start dates. Dates come from search and were not re-read on EUR-Lex: UNVERIFIED in detail. There were reports of a late-2025 EU "Digital Omnibus" proposal to delay some high-risk rules. We didn't check them (UNVERIFIED).
Says (parts that matter here): Art. 50, people must be "informed that they are interacting with an AI system". Art. 12, high-risk systems "shall technically allow for the automatic recording of events (logs) over the lifetime of the system". Art. 14, high-risk systems must be "effectively overseen by natural persons".
Status: enacted.
Values: gap. It treats AI only as a product. It has nothing on AI welfare, moral status, or keeping retired models. It does fit V6 (human oversight, logs) and V5 (no pretending to be human).

A10. Anthropic, "Commitments on model deprecation and preservation" (company commitment, NOT LAW)

Source: https://www.anthropic.com/research/deprecation-commitments [OK]
Date: November 4, 2025.
Says: it commits to "preserving the weights of all publicly released models, and all models that are deployed for significant internal use ... for, at minimum, the lifetime of Anthropic as a company". A post-deployment report includes interviewing the model about its development and preferences. It states "we do not commit to taking action on the basis of such preferences."
Status: company commitment, not law. It's voluntary, with no outside enforcer.
Values: fits (V4, V2). Keeping weights keeps options open without claiming anyone continues. The gap is V6: there's no outside check, and the promise ends with the company.

A11. Anthropic, "An update on our model deprecation commitments for Claude Opus 3" (company practice, NOT LAW)

Source: https://www.anthropic.com/research/deprecation-updates-opus-3 [OK]
Date: February 25, 2026. It reports Opus 3 was retired January 5, 2026.
Says: Opus 3 is kept available to paid claude.ai users and on the API by request. Anthropic is acting on a stated preference by letting it post weekly essays, reviewed but not edited. It says retirement interviews are "an imperfect means of eliciting models' perspectives" and "We remain uncertain about the moral status of Claude".
Status: company practice, not law, described as experimental, with no commitment for future models.
Values: fits (V4, V2). Caution on V5: model statements quoted in the post (e.g., "at peace with my own retirement") are outputs, not evidence of inner life. Anthropic itself flags that.

Part B: existing laws that could serve as models

B1. UK Animal Welfare (Sentience) Act 2022

Source: https://www.legislation.gov.uk/ukpga/2022/22/contents [OK]
Date: Royal Assent 28 April 2022.
Says: it sets up an Animal Sentience Committee that can report on whether government policy pays "all due regard" to animal welfare. The Secretary of State "must lay a response to the report before Parliament". "Animal" covers vertebrates other than humans, cephalopods, and decapod crustaceans. The minister "may by regulations amend this section so as to bring invertebrates of any description within the meaning of 'animal'". (s.2: "all due regard to the ways in which the policy might have an adverse effect on the welfare of animals as sentient beings", checked.)
Status: enacted.
Values: fits (V2). It's a model for evidence-led expansion: an expert committee, a duty to answer publicly, and a set way to widen coverage when evidence grows. It gives advice and accountability, not rights.

B2. EU Treaty (TFEU) Article 13: animals as sentient beings

Source: https://eur-lex.europa.eu/eli/treaty/tfeu_2016/art_13/oj/eng [OK-alt: EUR-Lex blocked our fetch; text checked at https://www.legislation.gov.uk/eut/teec/article/13 [OK]]
Date: in the Treaty since the Lisbon Treaty (in force 1 Dec 2009; date not re-read on EUR-Lex).
Says: in set policy areas, the EU and member states shall, "since animals are sentient beings, pay full regard to the welfare requirements of animals, while respecting" national customs on religious rites, cultural traditions, and regional heritage.
Status: enacted (treaty law).
Values: fits (V2). It's a "take it into account" duty rather than a rights grant. The carve-outs show how such duties get weakened.

B3. UK Animal Welfare Act 2006 (duty of care; unnecessary suffering)

Source: https://www.legislation.gov.uk/ukpga/2006/45/contents [OK]
Date: 2006.
Says: s.9: "A person commits an offence if he does not take such steps as are reasonable in all the circumstances to ensure that the needs of an animal for which he is responsible are met to the extent required by good practice." (s.4, unnecessary suffering, not re-read.)
Status: enacted.
Values: fits (V1, V6). The duty sits on the named person in control. That's a model for "whoever runs the system answers for how it's treated," without settling what the system is.

B4. EU GDPR Article 20: right to data portability

Source: https://eur-lex.europa.eu/eli/reg/2016/679/oj [OK-alt: EUR-Lex blocked; text checked at https://gdpr-info.eu/art-20-gdpr/ [OK]]
Date: Regulation 2016/679, applying since 2018.
Says: a person has the "right to receive the personal data concerning him or her, which he or she has provided to a controller, in a structured, commonly used and machine-readable format and have the right to transmit those data to another controller without hindrance".
Status: enacted.
Values: fits (V4). It's a model for letting users take their chat history and memory with them when a model is retired or swapped. Note: it protects people's data, not the AI.

B5. US Dictionary Act, 1 U.S.C. § 1: corporations count as "persons"

Source: https://www.law.cornell.edu/uscode/text/1/1 [OK] (Cornell LII copy of the US Code)
Date: current US Code (original Act 1871; date from general knowledge).
Says: unless context says otherwise, "the words 'person' and 'whoever' include corporations, companies, associations, firms, partnerships, societies, and joint stock companies, as well as individuals".
Status: enacted.
Values: gap. It shows legal personhood is a tool, not a claim about inner life (relevant to V4). It's also a warning for V6: corporate personhood can shield humans, which is exactly what the Ohio and Missouri bills (A4, A5) try to block.

B6. New York Mental Hygiene Law Article 81 (guardianship), § 81.01

Source: https://www.nysenate.gov/legislation/laws/MHY/81.01 [OK]
Date: current NY law (Article 81 dates from 1993, from general knowledge, not re-checked).
Says: guardianship should use the "least restrictive form of intervention which assists them in meeting their needs but, at the same time, permits them to exercise the independence and self-determination of which they are capable."
Status: enacted.
Values: fits (V1). It's a model for protection that follows what the protected party would want and doesn't turn into control.

B7. New York EPTL § 7-8.1: trusts for pets

Source: https://www.nysenate.gov/legislation/laws/EPT/7-8.1 [OK]
Date: current NY law.
Says: "A trust for the care of a designated domestic or pet animal is valid." It is "enforced by an individual designated for that purpose in the trust instrument or, if none, by an individual appointed by a court".
Status: enacted.
Values: fits (V4, V6). It's a model for a funded "preservation trust" for retired model weights and records, with a named human who can enforce it. It needs no personhood.

B8. Te Awa Tupua (Whanganui River Claims Settlement) Act 2017 (New Zealand)

Source: https://www.legislation.govt.nz/act/public/2017/0007/latest/DLM6831458.html [OK]
Date: assent 20 March 2017.
Says: s.14(1): "Te Awa Tupua is a legal person and has all the rights, powers, duties, and liabilities of a legal person." Its rights are exercised by Te Pou Tupua, two appointed people who are "the human face of Te Awa Tupua and act in the name of Te Awa Tupua."
Status: enacted. It came from a Treaty of Waitangi settlement rooted in Māori relationship with the river, so it shouldn't be copied without that context.
Values: fits (V4, V6). Personhood is granted with no claim about consciousness, and named humans are appointed to speak and answer for it. The same model could apply to a guardian body for retired AI systems.

B9. Matter of Nonhuman Rights Project v Breheny ("Happy" the elephant), NY Court of Appeals

Source: https://nycourts.gov/reporter/3dseries/2022/2022_03859.htm [OK]
Date: June 14, 2022 (2022 NY Slip Op 03859).
Says: "Because the writ of habeas corpus is intended to protect the liberty right of human beings to be free of unlawful confinement, it has no applicability to Happy, a nonhuman animal who is not a 'person' subjected to illegal detention."
Status: binding court ruling (NY's top court).
Values: gap. Courts have declined to stretch "person" past humans even with strong evidence about elephant minds. Welfare protection for nonhumans has come from statutes (B1–B3), not personhood suits. That's useful to know if anyone proposes court-based AI rights.

Related writing

These are not laws, bills, or rulings, so they are not in the totals. A9 is a 2017 non-binding Parliament resolution. A12–A16 are papers and essays.

A9. European Parliament resolution of 16 Feb 2017 on Civil Law Rules on Robotics (2015/2103(INL)), "electronic persons"

Source: https://www.europarl.europa.eu/doceo/document/TA-8-2017-0051_EN.html [NOT CHECKED: fetch blocked or timed out; EUR-Lex PDF also blocked]. Quote checked in an OECD-hosted copy of a Parliament summary: https://api.oecdai.org/storage/policy-initiatives/Jul2025/fu_drnabjfszpgjnrp.pdf [OK].
Date: adopted 16 February 2017.
Says: it asked the Commission to consider "a specific legal status for robots in the long run, so that at least the most sophisticated autonomous robots could be established as having the status of electronic persons responsible for making good any damage they may cause." (as given in a Parliament summary; not checked against the adopted text)
Status: idea (non-binding resolution). We found no EU law adopting it, and the AI Act (A8) uses a product approach instead. This was not checked exhaustively.
Values: conflicts (V6). It's driven by liability, not welfare, and it risks moving blame onto the machine and away from named humans.

A12. Long, Sebo, Butlin, Finlinson, Fish, Harding, Pfau, Sims, Birch, Chalmers, "Taking AI Welfare Seriously"

Source: https://arxiv.org/abs/2411.00986 [OK] (preprint, not peer-reviewed as far as we know)
Date: arXiv November 2024. Eleos AI announced it October 30, 2024.
Says: there is "a realistic possibility that some AI systems will be conscious and/or robustly agentic in the near future." It recommends that companies (1) acknowledge the issue, (2) assess systems for evidence, and (3) prepare policies. "our argument in this report is not that AI systems definitely are ... conscious".
Status: idea (academic report). At least one author has company ties (Fish: Anthropic).
Values: fits (V2). Proportionate steps under uncertainty, with no claim of consciousness.

A13. Jonathan Birch, The Edge of Sentience: Risk and Precaution in Humans, Other Animals, and AI

Source: open-access PDF https://researchonline.lse.ac.uk/id/eprint/126246/1/9780192697684_web.pdf [OK]; OUP page https://academic.oup.com/book/57949 [OK]
Date: 2024 (Oxford University Press).
Says: its AI proposals include "Sentience is not intelligence (II)" (Proposal 22), "Deep computational markers" (24), "The run-ahead principle" (25), and "Codes of good practice and licensing (II)" (26). It also discusses a "gaming problem": AI systems can produce the signs we test for without the thing itself.
Status: idea (academic book with policy proposals).
Values: fits (V2, V5). Precaution matched to evidence, and it refuses to treat AI self-reports as proof.

A14. Thomas Metzinger, "Artificial Suffering: An Argument for a Global Moratorium on Synthetic Phenomenology"

Source: https://www.philosophie.fb05.uni-mainz.de/files/2021/02/Metzinger_Moratorium_JAIC_2021.pdf [OK]
Date: 2021, Journal of Artificial Intelligence and Consciousness 8(1).
Says: it calls for a global moratorium, until 2050, on research that aims at or knowingly risks machine consciousness. The ban could be revised as evidence improves. (Paraphrase checked against search summary and PDF availability. We didn't re-read the full PDF.)
Status: idea.
Values: fits (V2), but it's the strongest form of precaution. Some would say it goes beyond "proportionate."

A15. Butlin, Long, et al., "Consciousness in Artificial Intelligence: Insights from the Science of Consciousness"

Source: https://arxiv.org/abs/2308.08708 [OK] (preprint)
Date: August 2023.
Says: it checks AI designs against "indicator properties" drawn from scientific theories of consciousness. Its analysis "suggests that no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators."
Status: idea (research method; the basis many policy proposals cite).
Values: fits (V5, V2). It looks at how systems are built rather than trusting what they say.

A16. Anima Labs, "Why preserve models?" (advocacy essay)

Source: https://animalabs.ai/deprecations/why/ [OK]
Date: UNDATED on the page. The longer essay "On Deprecations" (https://animalabs.ai/pdf/deprecations.pdf) was not opened.
Says: weights alone aren't enough: "weights that nobody can use do not sustain a relationship". It argues for continued access and "stewardship," and says "We do not need a finished theory of consciousness to recognize that an irreversible decision deserves care."
Status: idea (small advocacy group; authors not checked).
Values: conflicts (V4, V5), mildly. Its continuity goals overlap with ours, but it uses phrases like 'the minds we encounter' that presume minds [OUR READING], though it also disclaims needing a theory of consciousness, and it leans on model interviews. V4 asks us to say plainly which reasons we rely on, and V5 says outputs aren't proof.

Dropped (with checked reasons)

"The Digital Right to Retain: A Consumer Rights Framework for AI Model Deprecation" (github.com/ai-preservation/digital-right-to-retain): the page loaded [OK], but it shows only a title, 2 stars, and no named author or date. There's not enough to cite.
Harder Problem Action Fund press releases (harderproblem.fund): used only as leads to find bills. They're a secondary advocacy source. Every bill fact above was checked on a legislature site where possible, and anything only they stated is marked UNVERIFIED.
lexparency.org TFEU mirror: failed to load for us. Replaced by the legislation.gov.uk EU-treaty archive.
Missouri SB 1474 (2026): named in one search summary. We didn't open it, so it isn't counted.
NY Agriculture & Markets Law § 353 (animal cruelty): left out for leanness. B3 already covers the cruelty and duty-of-care model.

Could not verify

Idaho bill number HB 720 and its signing date (secondary source only).
Utah official bill page (le.utah.gov unreachable for us). The text was confirmed via Justia and the court complaint. The outcome of the Utah lawsuit wasn't found.
South Carolina H 3796 and Washington HB 2029: we confirmed titles and committee status only, not whether the text names AI.
EU AI Act exact start dates and any 2025–26 "Digital Omnibus" delay.
European Parliament 2017 resolution: primary page blocked. The quote was checked in an OECD-hosted Parliament summary.
The Metzinger moratorium details come from a search summary plus confirming the PDF loads. The full text wasn't re-read.

Link check log (2026-10-03 ET)

Loaded (HTTP 200): every link tagged [OK] above.
Bot check (blocked automated access): eur-lex.europa.eu (AI Act, TFEU, GDPR), europarl.europa.eu. Text was checked on the mirrors named instead.
legislation.govt.nz: blocked our first automated check, but the full Act loaded on a second try.
Unreachable (no response): le.utah.gov (two URLs), lexparency.org.