A G7 legislature rules out legal personhood for AI

Will the US Congress or the national legislature of another G7 country enact a statute declaring that AI systems cannot be legal persons?

Chance by Q4 2031

10%

Cumulative probability

0%25%50%75%100%
  • Latest forecast
Chance by the end of each quarter
YearQ1Q2Q3Q4
2026<1%
2027<1%<1%1%2%
20282%2%3%3%
20294%4%5%5%
20306%6%7%7%
20318%8%9%10%

Forecast history

Forecast made on 13 Oct 2026

Latest of 5 forecasts · Methodology v2.1

Model reasoning

I put the chance at about 1.5% by the end of 2027 and about 9.5% by the end of 2031. Four US states ban AI from being a legal person, but no such bill exists in Congress or any other G7 parliament. Even big federal AI bills leave the issue out, and Congress rarely passes AI laws. The October 2026 public fight over whether AI models like Claude might be conscious adds some upside. A bill with a sponsor that starts moving in Congress would change my forecast most.

Where things stand (13 Oct 2026)

No G7 national legislature has a bill to deny AI legal personhood, let alone passed one. All the activity is in US states, and state laws don't count for this question.

  • States. Four states have enacted bans: Idaho (2022), North Dakota (2023), Utah (2024) and Tennessee (SB 837, April 2026). Tennessee's is the first to expressly exclude AI from its definition of "person." About a dozen states have introduced such bills (source: AI Frontiers, 1 Sep 2026). Missouri's bill was killed in a House committee after pushback from business groups and worries about Trump's executive order on state AI laws. Oklahoma's stalled. Bills are expected in Ohio and Kentucky in 2027.
  • Congress. WIRED (14 Aug 2026) says "the federal government has failed to move on this issue."
    • The White House's March 2026 national AI legislative framework doesn't mention personhood. Neither does Sen. Blackburn's 300-page TRUMP AMERICA AI Act draft (I checked its section-by-section).
    • I read the GovInfo texts of the main 2026 AI bills: Sanders/Casar (S.5493/H.R.10538), the AI Emergency Button Act (S.5417), the FRONTIER Act (H.R.9925), H.R.9917, and the Warner bill (S.5576). I also read Rep. Trahan's CLAIM Act draft (7 Oct 2026) and the AI LEAD Act section-by-section. None has a personhood clause.
    • A GovInfo full-text search of 119th-Congress bills for "artificial intelligence" plus "personhood" returned nothing.
    • Congress is in recess until after the 3 Nov midterms. House Energy and Commerce Chair Guthrie says there will be no AI bill in the lame duck.
  • Other G7 countries. I found no relevant bills. Italy's Law 132/2025 and Japan's AI Promotion Act contain no such rule, Canada's old AI bill (AIDA) lapsed, and the UK's AI bill is still only under study (ActuIA overview, 5 Oct 2026).

Reference class

  • At the federal level the closest past cases are bioethics "status" bans with broad public support, such as the proposed human-cloning and human–animal chimera bans. Both were introduced repeatedly and never enacted.
  • State trends reach federal law slowly, if at all. The TAKE IT DOWN Act (2025) came about six years after the first state deepfake laws. There is still no federal privacy law and no KOSA.
  • The faster counter-example is the Defense of Marriage Act (1996), which passed within months of a perceived threat.
  • Federal AI statutes are rare. Even bipartisan bills like NO FAKES, KOSA and the GUARD Act have stalled.

Paths to YES

  1. A broad federal AI law (most likely 2027–2030) that includes a general "AI is not a person" clause.
    • The cross-party Pro-Human AI Declaration lists "No AI Personhood" among its principles. Its signers and allies run from Bannon and IFS to Sanders and teachers' and other unions, and Rep. Trahan spoke at its September assembly.
    • The Future of Life Institute's July 2026 model liability bill says a model "cannot be treated as a legal person."
    • Microsoft's 14 Sep 2026 guidelines say AI should not have legal personhood. One poll (June 2026) found 77% of voters say AI should always be a tool and only 11% favour rights for AI.
  2. A short stand-alone bill, or a rider on a must-pass bill.
    • The issue turned into a national culture-war fight in October 2026, over Anthropic's "don't be cruel to Claude" policy and its meetings with clergy. Pope Leo, DeSantis and Rep. Ted Lieu (who co-chairs House Democrats' AI commission: "math calculations… can never achieve personhood") all spoke against AI personhood.
    • Possible triggers include Delaware's planned AI-run companies and Milei's "non-human corporations" proposal in Argentina.
  3. Italy, France or another G7 country (about 2% over five years). The Pope's encyclical could influence Italy.

Paths to NO / strongest counterpoints

  • Even sympathetic conservative thinkers advise against personhood-ban language at the federal level. An IFS brief by Hawley's former chief counsel recommends product-liability and Section 230 fixes instead.
  • The Trump White House has fought state AI laws, and Congress is gridlocked.
  • A clause limited to one law ("for purposes of this Act"), to liability only, or to one sector would not count, and that is the most likely way such language would be drafted.

Forecast construction

  • I set a hazard for each quarter and computed the series in Python: 0.1% for Q4 2026 (lame duck, no bill exists), rising to about 0.35–0.55% per quarter in 2027–28, with small spikes at year-end packages, and 0.4–0.7% per quarter in 2029–31.
  • The hazards sum to 10%, giving cumulative probabilities of 1.5% by end-2027, 3.3% by end-2028 and 9.5% by end-2031. Roughly 7.5 points of that come from the US and 2 from the other G7 countries.
  • Case for higher: a fast, DOMA-style reaction to a trigger event, now that there is bipartisan rhetoric. I raised the later hazards for this, but there is still no bill and no champion in Congress, so I hold below 10%.

Resolution criteria

Question. By the last day of each quarter, has the US Congress or the national legislature of another G7 country enacted a statute declaring that AI systems cannot be legal persons?

Definitions.

  • G7 country: Canada, France, Germany, Italy, Japan, the United Kingdom, or the United States. EU legislation does not itself count.
  • Enacted statute: primary legislation passed by the national legislature and completed through the country's required assent, signature, promulgation, veto override, or other constitutional enactment procedure. Legislative passage alone is insufficient when further steps are required. A future commencement date does not prevent an enacted statute from counting.
  • AI systems: artificial-intelligence software, models, or AI-equipped machines expressly identified in the statutory text, including through a definition incorporated by reference. The statute's definition controls; there is no restriction to covered developers or frontier systems.
  • Cannot be legal persons: an express, legally operative rule that AI systems, as a class, are not persons or cannot be granted legal personality in their own right under national law generally. Equivalent language denying AI systems independent legal subjecthood counts. A nationwide federal-law prohibition counts even if the statute does not preempt state or provincial law. A prohibition may be temporary or subject to future legislative amendment.

Resolves YES if an official source publishes the enacted text and confirms completion of enactment, and the statute contains the qualifying prohibition. The prohibition may appear in a standalone act, an amendment, or a larger statute. A generally applicable definition of “person” expressly excluding AI counts. Denial of personhood to other entities in the same provision does not disqualify it.

Resolves NO / Exclusions.

  • No qualifying official publication by the quarter-end.
  • Bills, passage by only one chamber, uncompleted enactment, vetoed bills without a completed override, nonbinding resolutions, legislative findings or preambles without an operative prohibition, and statements in hearings or explanatory memoranda.
  • Executive orders, administrative rules, judicial decisions, treaties without qualifying implementing legislation, and subnational statutes. For example, Idaho H0720 (2022) and Utah HB0249 (2024) do not count.
  • Merely leaving AI unrecognized under existing law, requiring human oversight, assigning liability to humans or companies, declaring AI non-conscious, or denying AI a particular right or role, such as authorship, inventorship, voting, or public office.
  • Excluding AI from “person” solely for a particular sector, offense, program, or regulatory regime, rather than denying legal personality generally. A ban applying only to a named system or particular application also does not count.
  • The legal personality of a corporation owned, managed, or operated by AI is not the personality of the AI itself.

Resolving source. Official enacted statutory text and official enactment records only, a narrower source rule than Public confirmation (A2): United States—Congress.gov or GovInfo Public and Private Laws; Canada—Parliament's LEGISinfo and Justice Laws Website Annual Statutes or Canada Gazette Part III; United Kingdom—legislation.gov.uk Acts as enacted and Parliament's Royal Assent records; France—Journal officiel de la République française on Légifrance; Germany—Bundesgesetzblatt on recht.bund.de; Italy—Gazzetta Ufficiale della Repubblica Italiana; Japan—Kanpō, with the Diet's enactment records or e-Gov Laws statutory text as corroboration. Official successor repositories count. Apply the Resolution window and Resolution date (A3): use the first qualifying publication on or after 2026-09-16 establishing both the enacted text and completed enactment, not the commencement date. Republishing an unchanged pre-opening statute does not constitute a new enactment. Later repeal, expiry, or invalidation does not reverse a YES.

Download all forecasts (JSON)