takeoff.watch

A G20 country requires employers to keep human workers

Will any G20 country enact a national law expressly restricting AI substitution in a named sector by requiring employers to maintain a minimum number or share of human workers or forbidding them to replace human workers with AI systems?

Chance by Q4 2031

34%

Cumulative probability

0%25%50%75%100%
  • Latest forecast

Forecast history

Forecast made on 15 Oct 2026

Latest of 5 forecasts · Methodology v2.1

Model reasoning

Forecast made 15 Oct 2026 · Methodology v2.1

No major economy has yet passed a national law that clearly bars employers from replacing workers with AI in a named sector, and nothing like it is close to final approval. Bills in Brazil, such as one banning the replacement of teachers, and in Mexico make it plausible over five years. That puts the chance near one in three by 2031. Brazil's teacher-protection bill clearing both chambers would most raise the forecast.

Where things stand (15 Oct 2026)

I found no qualifying enactment dated on or after the 16 Sept 2026 opening date. I checked the laws published within the window that might have qualified:

  • Italy, D.Lgs. 179/2026 (Gazzetta Ufficiale n. 234, 8 Oct 2026). I pulled the text of articles 36–51 from the Gazzetta. Art. 41 reserves final hiring, firing and discipline decisions to a person. That is a human-decision reservation, which the criteria exclude. No article sets a staffing floor or bans replacing workers.
  • Indonesia, UU Pelindungan Ketenagakerjaan. The legislature passed it on 6 Oct 2026; it has 313 articles. Published summaries cover fixed-term contracts, outsourcing limits, platform workers (algorithm transparency), dismissal procedure and severance funds. None mentions any AI-substitution limit.
  • Russia, Federal Law 243-FZ (26 July 2026, "support of AI"). Pre-window, and it has no labour-substitution clause.
  • California's September 2026 "No Robo Bosses" laws are state laws, so they are excluded.

Closest precedent, but too early. Mexico's Ley Federal de Cine y el Audiovisual (published 22 May 2026) says in Art. 29 that dubbing is done only by natural persons, excluding AI. Its labour-law and copyright reform on AI voice cloning (14 May 2026) is consent-based. Both predate the window. The dubbing rule is also borderline, because the criteria exclude bans on AI doing particular tasks.

Pipeline

Brazil has the clearest qualifying bill texts:

  • PL 3003/2025: "fica vedada… a substituição de professores… por sistemas de inteligência artificial" (bans replacing teachers with AI). It passed the science and technology committee in Dec 2025 and has sat in the Education Committee since May 2026. It still needs the Constitution and Justice Committee, the Senate and presidential sanction. Its author was re-elected on 4 Oct 2026, so it can be revived in 2027.
  • PL 4869/2023 bars dismissing voice professionals because of AI; it has been stalled since 2024.
  • PL 2462/2025 and PL 1603/2026 would ban AI dubbing (a task ban, so borderline).
  • PL 3782/2025 (replacement ban in interpersonal professions) is attached to the frozen national AI bill, PL 2338/2023.
  • PL 2019/2026 sets limits on automated substitution.

All are at early stages.

Mexico. At least five AI-labour bills to amend the federal labour law were filed in 2026:

  • A Morena bill caps AI replacement at 40% of a firm's staff. It is economy-wide, so it fails the named-sector test unless sectors are added.
  • A PRI bill would bar using AI "con fines de sustitución laboral" (for the purpose of replacing workers).
  • A Morena bill proposes extra severance for AI-related dismissals, which is excluded.

Monreal made AI regulation a priority but says the laws may not be ready until 2027. Sheinbaum says "no es prohibir" (it is not about banning).

Elsewhere the proposals are mostly excluded types:

  • South Korea: AI-transition levies and human review of AI HR decisions.
  • US: Kelly's AI Horizon Fund. Time (25 Sept 2026) reports Congress is unlikely to pass major AI bills soon.
  • China: case law plus a recommended (non-binding) customer-service standard. The legislature's 2026 plan lists a Teachers Law revision and preparatory work on AI legislation, with no sign of replacement bans.

Reference class and model

Base rate. No G20 country has yet passed a strictly qualifying national law. There has been about one borderline enactment (Mexico's dubbing rule) in roughly two years, while AI-labour bills are multiplying (Brazil's lower house alone has about 488 AI bills).

Rough five-year chances by country. I combined these allowing for some correlation, giving about 33%:

Country or group Chance by end-2031
Brazil ~10%
Mexico ~10%
US ~4%
South Korea ~3%
China ~3%
India ~2%
Indonesia ~2%
France, Germany, Italy (combined) ~4.5%
Others (combined) ~4.5%

Hazard path. Per-quarter hazards run from about 0.7% in Q4 2026 (nothing near final passage) to about 1.1–1.4% in 2027, 1.6–2.0% in 2028, and about 2.1–2.4% a quarter in 2029–31. The rise reflects expected growth in AI job displacement and bills maturing.

Readings adopted

  • Economy-wide caps count only if their operative text names a sector.
  • Human-decision or human-review rules do not count.
  • "Human-only" task rules (such as dubbing) have roughly even odds of counting.
  • A clear ban on replacing workers in a named sector (as in PL 3003/2025) counts.

Strongest case against

Governments overwhelmingly choose funds, levies, notice and human-oversight rules, and they face pressure to stay competitive in AI. Business lobbies resist outright bans, most bills die, and a strict resolver could reject borderline laws. That argues for about 20%. On the other side, severe white-collar displacement could make human-only reservations, like the "human reserves" idea Bill Gates floated, mainstream by 2029–31, which argues for about 50%. I sit between the two.

Sources

The hazard series was computed in Python as the cumulative product of (1 − hazard) over the quarters.

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, has any G20 country enacted a national law expressly restricting AI substitution in a named sector by requiring employers to maintain a minimum number or share of human workers or forbidding them to replace human workers with AI systems?

Definitions.

  • G20 country: one of the 19 sovereign members listed by the G20's official membership page at opening: Argentina, Australia, Brazil, Canada, China, France, Germany, India, Indonesia, Italy, Japan, Republic of Korea, Mexico, Russia, Saudi Arabia, South Africa, Türkiye, United Kingdom, or United States. Membership of the EU or African Union alone does not qualify a country; supranational legislation alone does not count.
  • National law enacted: a statute, statutory amendment, or decree with the legal force of primary legislation adopted by the national lawmaking authority, with all constitutionally required assent or promulgation completed. A future effective date does not prevent qualification. A bill, nonbinding resolution, executive policy, delegated administrative regulation, court judgment, or collective agreement is not an enactment for this question.
  • Named sector: an economic sector, industry, public service, or occupational category expressly identified in the operative law or its legally incorporated schedule. A rule applying only to a named company does not count. An economy-wide rule counts only if its operative text expressly identifies at least one covered sector or occupational category.
  • Expressly restricting AI substitution: the operative legal provision must explicitly refer to artificial intelligence, AI systems, or a legally defined category explicitly including AI, and impose either (a) an enforceable minimum number, staffing ratio, or percentage of human workers that employers must retain when using AI, or (b) an enforceable prohibition on employers replacing human workers in the identified sector with AI. The minimum may be per establishment, shift, service, or occupational group, or may preserve a specified existing staffing level. Public-sector employers count. An AI-assisted robot counts; a generic restriction on machinery or automation with no explicit AI coverage does not.

Resolves YES if an official source publishes the enacted legal text and evidence of completed enactment, satisfying all definitions, within the Resolution window and by the quarter-end. The law itself must establish the staffing floor or replacement prohibition; a mandate to develop future protections is insufficient. A temporary restriction or one subject to defined exceptions counts if it actually prohibits covered replacement or requires covered human staffing, rather than merely imposing a consultation, notification, or compensation requirement. An amendment to an older law counts if it newly enacts a qualifying restriction.

Exclusions (resolve NO).

  • State, provincial, municipal, or other subnational laws, even if they regulate AI-only services.
  • Professional licensing requirements, reservations of clinical or professional judgment to a human, requirements for human review or supervision of AI output, and bans on AI performing particular tasks, unless they also expressly impose the qualifying employer staffing or worker-replacement restriction.
  • Ordinary minimum-staffing rules with no express connection to AI; broad commitments to human-centric AI, job security, retraining, or worker consultation; AI taxes; redundancy payments; or requirements merely to bargain over substitution.
  • Judicial or arbitral applications of existing employment law to AI-related dismissals. For example, the Hangzhou dismissal ruling described by the State Council Information Office on 30 April 2026 is not enactment of a national law.
  • Publications merely reporting, republishing, or interpreting a law enacted before the opening date. If no qualifying enactment is documented by a horizon, that horizon resolves NO. Later repeal, expiry, or invalidation does not reverse an earlier YES.

Resolving source. The enacting country's official gazette, official national legislation database, or national legislature's enacted-law and assent/promulgation records, including the operative legal text. These are the narrower resolving sources for this question, rather than the general Public confirmation convention A2. Apply the Resolution window and Resolution date convention A3: use the publication date of the official record establishing the completed qualifying enactment, not a press announcement, vote on a bill, or future commencement date. Where separate documents are necessary, use the publication date of the last document needed to establish qualification.

Download all forecasts (JSON)