Scan: U.S. enacts AI safety bill before 2027?

1. Planner Fable 5.1 · Fermi decomposition + tool plan

Tools chosen: polymarket_direct congress_bills congress_bills congress_bills claude_news article_search gdelt_news polymarket_related kalshi_related wikipedia

Sub-questions (Fermi decomposition)

What is the current Polymarket YES price and how has it trended over the past 90 days?
Have any bills with AI usage restrictions or human-in-the-loop provisions already been signed into law in the 119th Congress (e.g., TAKE IT DOWN Act, NDAA FY2026 AI provisions), and would they satisfy the resolution criteria?
Which pending federal AI bills (e.g., NO FAKES Act, CHAT Act, GUARD Act, AI LEAD Act, Kids Online Safety Act with AI provisions, nuclear launch human-in-the-loop bills) have passed at least one chamber or been marked up, and what is their trajectory?
Does the FY2027 NDAA or any must-pass appropriations vehicle include AI provisions (e.g., human oversight of autonomous weapons, restrictions on Chinese AI models like DeepSeek in federal systems) that would qualify?
What is the Trump administration's and Republican congressional leadership's stated position on federal AI regulation versus preemption of state laws, and does the December 2025 AI preemption executive order affect legislative prospects?
How is the Polymarket community interpreting the resolution criteria—does a narrow usage restriction (e.g., banning federal agency use of specific foreign AI models) count?
What is the base rate of Congress enacting technology-regulation legislation within a two-year window, and how many bills have been introduced vs. enacted?

Planner reasoning

This is a Polymarket question about whether any federal bill containing AI safety provisions (prohibitions, training limits, usage restrictions, or human-in-the-loop requirements) becomes law by end of 2026. Key uncertainties include the status of specific bills (e.g., TAKE IT DOWN Act already enacted in 2025 may or may not count, NDAA AI provisions, AI preemption/moratorium efforts, kids' AI chatbot bills), the broad interpretation of the resolution criteria, and the current administration's deregulatory stance versus bipartisan momentum on narrow AI harms. Research should anchor on the Polymarket price, check Congress.gov for pending AI legislation, and scan news for legislative momentum and resolution-criteria debates.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.3s	1	## This Market's Polymarket Data U.S. enacts AI safety bill before 2027? - Current price (probability): 34.50% - 7-day price change: +25.50% - 30-day price change: +22.50% - Total volume: $104,574 (USD notional) - Price range: 9.00% - 43.50% - Data points: 91 days
congress_bills	OK	3.6s	18	Found 18 bills matching 'artificial intelligence'.
congress_bills	OK	3.4s	9	Found 9 bills matching 'AI chatbot minors'.
congress_bills	OK	3.7s	8	Found 8 bills matching 'autonomous weapons human control nuclear'.
claude_news	OK	23.9s	11	## Key Findings: Federal AI Safety Legislation Status (as of September 2026) - No comprehensive federal AI safety act exists yet. As of September 2026, the U.S. still has no federal AI Act; the landscape maps to EO 14365/14409, plus state laws like California SB 53, Texas TRAIGA, and New York
article_search	OK	4.6s	25	Fetched 25 unique articles across 4 queries (lookback=60d). 'Congress AI safety bill vote 2026': 10 articles \| 'AI chatbot children legislation Senate': 10 articles \| 'NDAA artificial intelligence provisions signed': 10 articles \| 'federal AI preemption state laws moratorium Congress': 10 articles
gdelt_news	OK	170.5s	20	GDELT: 20 articles across 3 queries (lookback=30d). 'AI regulation bill Congress': error GDELT rate-limited after retries (429) \| 'AI safety legislation signed into law': 10 hits \| 'GUARD Act AI chatbots': 10 hits
polymarket_related	OK	2.5s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'AI regulation': 0 markets \| keyword 'AI bill': 0 markets \| keyword 'artificial intelligence law': 0 markets
kalshi_related	OK	2.4s	1	1 related markets / summaries. keyword 'AI regulation': no matches \| keyword 'AI bill': ok \| keyword 'artificial intelligence': no matches
wikipedia	OK	0.2s	3	Fetched 3 Wikipedia entries (0 missing pages).

3. Evidence Brief Sonnet 5 · 8989 chars

# Current state No federal law meeting the resolution criteria (AI creation/training/usage bans or human-in-the-loop mandates) has been enacted as of mid-September 2026. The two closest enacted measures — TAKE IT DOWN Act (May 2025, deepfake NCII ban) and NDAA FY2026's DeepSeek/China AI procurement ban (Dec 2025) — are narrow and of ambiguous qualification. A bipartisan Klobuchar-Cruz-Thune AI safety bill is reportedly being drafted (Sept 2026) but has no confirmed text or floor schedule, and the House just recessed until after November midterms, compressing the runway to ~6-7 weeks before the Dec 31, 2026 deadline. # Timeline of key events - 2025-05-19: TAKE IT DOWN Act signed into law (deepfake NCII ban); enforcement rules effective 2026-05-19. [confirmed, Wikipedia/legisletter] - 2025-09-29: AI LEAD Act (Durbin) referred to Senate Judiciary. [confirmed, congress.gov] - 2025-12-11: Trump signs EO 14365 "National Policy Framework for AI" — federal preemption push via executive action, not legislation. [confirmed, Wikipedia] - 2025-12-18: FY2026 NDAA (P.L. 119-60) signed; includes DeepSeek/High Flyer exclusion for DoD contractors — narrow procurement rule. [confirmed, governmentcontractslegalforum.com] - 2026-01-09/10: DOJ AI Litigation Task Force stood up to challenge state (not federal) AI laws. [confirmed, introl.com] - 2026-05-11: GUARD Act (Hawley) placed on Senate Legislative Calendar. [confirmed, congress.gov] - 2026-06 (follow-up): EO 14409 issued, extending national AI policy framework. [reported, CDT via claude_news] - 2026-06-23: SAFE KIDS Act introduced (Curtis). [confirmed, congress.gov] - 2026-07: FRONTIER Act (Obernolte/Trahan) and AI Kill Switch Act (Lieu/Moran) introduced; latter follows OpenAI-Hugging Face hack. [confirmed, CNBC] - 2026-08-05: KOSA + Youth AI Privacy Act advance out of Senate Commerce Committee (with AI chatbot safeguard provisions). [confirmed, CNBC/congress.gov] - 2026-07-27: FY2027 NDAA (S.4784) motion to proceed in Senate — no confirmed AI human-in-the-loop text yet. [reported, congress.gov] - 2026-09 (early): Sanders/Casar introduce Ban Artificial Superintelligence Act. [confirmed, CNBC] - 2026-09-10: Klobuchar-Cruz-Thune bipartisan AI safety bill reported "in the works," possible introduction imminent. [rumored/reported, Semafor] - 2026-09-13: CNBC reports Congress under pressure to act before midterms; House recessing until November, Democrats demand return to session. [confirmed, CNBC] # Event Will the U.S. enact a federal AI safety bill (with qualifying prohibition/training/usage/human-in-the-loop provisions) into law before Jan 1, 2027? # Outcomes to forecast Yes / No # Kalshi market anchor No direct Kalshi price returned by kalshi_direct tool in this research; only related/unrelated Kalshi markets surfaced (no AI-specific match). **Use Polymarket as best available cross-market proxy**: current YES 34.5%, up sharply from a 90-day low of 9% (+22.5% 30-day, +25.5% 7-day), on modest volume (~$105K total). This is a fast-moving, low-liquidity market — treat with caution. # Sub-question answers 1. **Polymarket trend** — Current 34.5%, range 9%-43.5% over 91 days; sharp recent rise (+22.5%/30d) likely tied to Sept 2026 news of Klobuchar-Cruz-Thune bill and heightened AI-safety-crisis coverage (CNBC). [polymarket_direct] 2. **Already-signed bills** — TAKE IT DOWN Act (May 2025) bans AI deepfake NCII creation/distribution — arguable but narrow "creation/release" fit, not clearly a broad "AI safety bill." NDAA FY2026 (Dec 2025) bans DeepSeek/China AI use by DoD contractors — a usage restriction but scoped to federal contracting, not general AI systems. Neither is confirmed by resolution source as satisfying criteria; ambiguous. [claude_news, Wikipedia] 3. **Pending bills' trajectory** — GUARD Act (Senate, on calendar since May 2026), CHAT Act 2.0 (introduced July 2026), KOSA + Youth AI Privacy Act (ordered reported from Senate Commerce Aug 2026), AI LEAD Act (referred Sept 2025, no floor action), FRONTIER Act and AI Kill Switch Act (both introduced July 2026, no committee action reported), Ban Artificial Superintelligence Act (introduced Sept 2026). None have passed both chambers; several sit at committee/calendar stage only. [congress_bills, CNBC] 4. **NDAA/appropriations vehicle** — FY2026 NDAA already law with narrow DeepSeek ban. FY2027 NDAA (S.4784) in Senate floor process (motion to proceed, July 2026) but no confirmed AI human-in-the-loop or model-restriction language reported; earlier attempts to insert AI preemption language into NDAA reportedly failed. [congress_bills, lw.com] 5. **Administration/GOP leadership stance** — Trump favors federal preemption of state AI laws (EO 14365 Dec 2025, EO 14409 June 2026) and DOJ litigation against state laws, not new restrictive federal legislation; GOP is split (GUARDRAILS Act/States' Right to Regulate AI Act vs. Baumgartner's preemption-codifying bill). This favors preemption/deregulation, working against passage of a restrictive "AI safety bill." [claude_news, Wikipedia] 6. **Polymarket community interpretation** — No direct evidence found on nuanced resolution interpretation; narrow procurement bans (DeepSeek) do not appear to have moved Polymarket price at time of enactment (Dec 2025), suggesting the market treats them as insufficient. [inferred, gap] 7. **Base rate** — Not directly quantified in research, but historical pattern shown: KOSA has failed to become law across multiple Congresses (2022-2024) despite Senate passage; only one AI-specific bill (TAKE IT DOWN) reached enactment in ~2 years of the current AI legislative surge out of ~18+ AI bills tracked — suggesting a low enactment rate for comprehensive AI bills. [Wikipedia, congress_bills] # Key facts (high-confidence, factual) 1. [Wikipedia] TAKE IT DOWN Act signed May 19, 2025 — only clearly enacted AI-adjacent federal law to date. 2. [governmentcontractslegalforum.com] FY2026 NDAA (P.L.119-60) signed Dec 18, 2025, includes narrow DeepSeek/China AI procurement ban. 3. [Wikipedia/claude_news] Trump signed preemption EO 14365 (Dec 2025) and follow-up EO 14409 (June 2026) — executive action, not legislation. 4. [Semafor] Klobuchar-Cruz-Thune bipartisan AI safety bill in drafting as of Sept 10, 2026; no confirmed text/schedule. 5. [CNBC] House recessing until after Nov 2026 midterms, compressing legislative window before Dec 31, 2026 deadline. 6. [congress_bills] 18+ AI-related bills introduced in 119th Congress; none besides TAKE IT DOWN/NDAA provisions have become law. # Cross-market signals - Kalshi related: No direct match found; unrelated markets only (Belichick, Kai Cenat, reconciliation bill). - Polymarket: 34.5% YES, sharp recent uptrend, thin volume (~$105K) — likely reflects recent news momentum more than settled consensus. - Sportsbook implied: N/A (not applicable to legislative event). # Analyst opinions and speculation - Brad Carson (Public First, pro-safety PAC) predicts "major AI legislation within the next eight months" as of Aug 2026 — optimistic outlier. [CNN] - CNBC (Sept 13, 2026) frames legislative window as "closing" given midterm recess, with Democrats pressuring Johnson to stay in session — implies urgency but no guaranteed action. - OpenAI publicly lobbying for "mandatory national AI safety rules" (Sept 2026), indicating industry now supportive of some federal action, which could accelerate passage. # Directional lean per outcome - **Yes**: Growing bipartisan momentum (Klobuchar-Cruz-Thune, GUARD Act, KOSA, Kill Switch Act), industry now pushing for rules, several bills at advanced committee/calendar stages, Thune's direct involvement as Majority Leader. - **No** (favored): No bill has passed both chambers; House out until November; historical pattern (KOSA multi-year failure) shows chamber disagreement is persistent; Trump administration favors preemption/deregulation over new restrictions; narrow enacted measures (TAKE IT DOWN, NDAA DeepSeek ban) may not satisfy strict resolution criteria per Polymarket's continued sub-50% pricing even after enactment. # Gaps / unknowns - No live Kalshi YES price was returned for this specific ticker — anchor is Polymarket only. - Unclear whether resolution source would count NDAA FY2026 DeepSeek ban or TAKE IT DOWN Act as satisfying criteria (ambiguity noted by market not spiking on these events). - No confirmed text of Klobuchar-Cruz-Thune bill; unknown whether it would include qualifying provisions. - FY2027 NDAA final text/AI provisions unconfirmed as of research cutoff. # Calibration anchors - Polymarket YES 34.5% (recent range 9%-43.5%, highly volatile, low liquidity) — primary cross-market anchor. - Precedent: KOSA introduced 2022, still not law by Sept 2026 despite Senate passage — illustrates multi-year failure pattern for even bipartisan-favored tech bills. - Only 1 of ~18+ tracked federal AI bills enacted in ~2 years of intense legislative activity — supports base-rate skepticism toward Yes.

4. Forecasters Fable 5.1 + GPT-5.6 Sol · in parallel

claude-fable-5-1

45% conf

Yes	30%
No	70%

No qualifying federal AI safety law has been enacted as of mid-September 2026, no pending bill has passed both chambers, and the House is recessed until after the November midterms, leaving only a compressed lame-duck window before Dec 31. The base rate is poor (1 of 18+ tracked AI bills enacted, KOSA failing across multiple Congresses) and the administration favors preemption/deregulation over new restrictions. Upside risk comes from the Klobuchar-Cruz-Thune draft with Majority Leader involvement, industry (OpenAI) now lobbying for federal rules, and the FY2027 NDAA as a must-pass vehicle that could carry human-in-the-loop or model-restriction language, plus resolution ambiguity around already-enacted narrow measures. With no direct Kalshi price, I anchor on Polymarket's 34.5% but shade slightly lower given the thin, news-driven rally and the tight calendar.

gpt-5.6-sol

67% conf

Yes	28%
No	72%

The 34.5% Polymarket YES price is the best available anchor, but its sharp rise on modest liquidity appears driven by news of a still-unwritten Klobuchar-Cruz-Thune proposal. Pending bills show bipartisan interest, yet none has passed both chambers, the House recess leaves only roughly 6–7 legislative weeks, and the FY2027 NDAA has no confirmed qualifying AI provision. The administration’s preference for preemption and deregulation, combined with KOSA’s repeated failure and the low enactment rate among more than 18 AI bills, points toward No. Existing TAKE IT DOWN and DeepSeek restrictions add resolution ambiguity, but market behavior suggests they are unlikely to qualify.

5. Devil's Advocate Sonnet 5 · red-team critique

Both forecasts converge tightly (0.28–0.30 Yes) and shade below the Polymarket 34.5% anchor for similar reasons, but neither explicitly quantifies how much weight to give a thin, volatile, low-liquidity ($105K) market that swung from 9% to 43.5% in 90 days — they should have discussed more explicitly why they discount it by only ~5-6 points rather than more given the brief's warning to "treat with caution."

Neither forecaster seriously engages with the lame-duck/NDAA vehicle scenario as a distinct pathway: FY2027 NDAA (S.4784) is a must-pass bill already in Senate floor process, and past NDAAs have carried narrow AI provisions (FY2026's DeepSeek ban) — a last-minute rider with human-in-the-loop language attached to a must-pass vehicle in December is a plausible non-trivial path to Yes that both mention only in passing without adjusting probability mass for it.

The resolution ambiguity sub-question (whether TAKE IT DOWN Act or NDAA DeepSeek ban already qualify) is treated as resolved toward "No" based on inferred market behavior ("market not spiking on these events"), but this is a weak inference from a thin market and not a confirmed resolution-source ruling — both forecasters could be underweighting the tail scenario where a resolver retroactively counts an existing narrow law as qualifying, resulting in an already-locked-in Yes.

Both analysts anchor heavily on the "1 of 18+ AI bills enacted" base rate without adjusting for the fact that legislative activity and bipartisan urgency have visibly accelerated in 2026 (GUARD Act, KOSA, Kill Switch Act, Klobuchar-Cruz-Thune all advancing within months of each other) — this base rate may be reference-class-stale relative to the current, unusually compressed burst of momentum described in the brief's Sept 2026 "pressure to act before midterms" framing.

Neither forecast addresses the Brad Carson analyst outlier prediction ("major AI legislation within 8 months" as of Aug 2026) or OpenAI's public lobbying shift, beyond a cursory mention — these represent a potential inflection point (industry no longer opposing regulation) that could materially change legislative dynamics versus the historical KOSA-style gridlock, and deserved more explicit discounting or weighting rather than being folded silently into "upside risk."

Both forecasts land at nearly identical probabilities (0.30 vs 0.28) despite Forecaster 2 expressing much higher confidence (0.67 vs 0.45) — this divergence in stated confidence without a corresponding difference in probability or rationale depth is inconsistent and neither explains what evidence justifies the higher certainty.

6. Aggregator Fable 5.1 · final calibrated forecast

30% (56% confidence)

Yes: 30% No: 70%

Both forecasters converge near 0.28–0.30, slightly below the thin Polymarket anchor of 34.5%, and the critique's points largely offset each other: the low-liquidity, news-driven rally argues for discounting the market further, while the FY2027 NDAA must-pass vehicle and unresolved ambiguity over whether TAKE IT DOWN or the DeepSeek procurement ban already qualify argue for keeping meaningful Yes mass. No qualifying bill has passed both chambers, the House is out until after midterms leaving a ~6–7 week lame-duck window, and the administration favors preemption over new restrictions, so No remains clearly favored. However, Majority Leader Thune's direct involvement, OpenAI's lobbying shift, and the precedent of narrow AI riders on the NDAA make Yes a real tail rather than a negligible one. I land at 0.30 Yes, essentially agreeing with the consensus and a modest shade below the market.

Pipeline Timing

Total pipeline time: 278.9s

Per-tool research timings shown in the Research section above.