Scan: US Government removes public access to a major Chinese AI model in 2026?

1. Planner Fable 5.1 · Fermi decomposition + tool plan

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

Sub-questions (Fermi decomposition)

What is the current Polymarket YES price and how has it trended since market creation?
Which bills in the 119th Congress would ban or restrict DeepSeek or other Chinese AI models for the general public (not just government devices), and how far have they advanced?
Has the Commerce Department (BIS/ICTS), FCC, or White House signaled or initiated any rulemaking, executive order, or ICTS supply-chain action targeting Chinese AI models like DeepSeek for US consumers?
What is the current status of US-China trade/tech relations and the Trump administration's stance on Chinese AI apps (e.g., DeepSeek app store bans, state-level bans, TikTok precedent)?
Have any states or federal agencies already banned DeepSeek, and is there momentum toward a nationwide public ban comparable to the TikTok divest-or-ban law?
How quickly could a formal action actually take effect (remove access) before Dec 31, 2026, given typical rulemaking and legislative timelines?
Are there related Kalshi or Polymarket markets on DeepSeek/Chinese AI bans or US-China tech restrictions that give an independent probability signal?

Planner reasoning

This is a Polymarket policy question about whether the US government will take formal action (legislation, EO, export control, Commerce/BIS rule, FCC action, etc.) that effectively removes general US public access to a flagship model from DeepSeek, Alibaba, ByteDance, etc. by end of 2026. Key drivers are pending legislation (e.g., No DeepSeek on Government Devices Act, broader bans), Commerce Department ICTS authority actions, and administration posture on Chinese AI. The market price is the primary anchor; legislative tracking, news, and cross-venue signals will refine it.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.5s	1	## This Market's Polymarket Data US Government removes public access to a major Chinese AI model in 2026? - Current price (probability): 20.50% - 7-day price change: +7.00% - 30-day price change: +8.00% - Total volume: $44,250 (USD notional) - Price range: 7.50% - 28.50% - Data points: 67 days
polymarket_related	OK	3.8s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'DeepSeek': 0 markets \| keyword 'Chinese AI': 0 markets \| keyword 'China ban': 0 markets \| keyword 'export control': 0 markets
kalshi_related	OK	3.4s	2	2 related markets / summaries. keyword 'DeepSeek': no matches \| keyword 'Chinese AI ban': ok \| keyword 'China export controls': ok
congress_bills	OK	3.2s	10	Found 10 bills matching 'DeepSeek'.
congress_bills	OK	1.1s	6	Found 6 bills matching 'Chinese artificial intelligence ban'.
article_search	OK	0.8s	37	Fetched 37 unique articles across 4 queries (lookback=60d). 'US ban DeepSeek public access': 10 articles \| 'Commerce Department Chinese AI model restriction': 10 articles \| 'executive order Chinese AI DeepSeek Qwen': 10 articles \| 'BIS export control AI model China app': 10 articles
claude_news	OK	22.1s	8	Based on my research (out of tool calls now), here are the key findings on 2026 US government actions regarding Chinese AI models: - Government device bans remain the norm; no consumer/public-facing ban exists. As of Sept 2026, DeepSeek is already forbidden in New York, Texas, and Virginia , a
gdelt_news	OK	132.7s	20	GDELT: 20 articles across 3 queries (lookback=45d). 'DeepSeek ban United States': error GDELT rate-limited after retries (429) \| 'Chinese AI model export control US': 10 hits \| 'Trump administration Chinese AI apps': 10 hits
wikipedia	OK	0.2s	3	Fetched 3 Wikipedia entries (0 missing pages).

3. Evidence Brief Sonnet 5 · 8574 chars

# Current state As of Sept 2026, no US federal action has removed general public access to any major Chinese AI model. All confirmed restrictions (Commerce Dept bureaus, Navy, NASA, states NY/TX/VA) are device/network-level bans on government systems, not public bans. Federal legislation targeting broader restrictions (HR1121/S765 "No DeepSeek on Government Devices Act") remains stalled in committee and is scoped only to government IT, not public access. Momentum toward broader China-tech restrictions (robots, inverters, chip distillation sanctions) is rising but has not yet touched consumer AI model access. # Timeline of key events - 2025-01: Commerce Dept blacklists Zhipu AI (Z.ai) on Entity List over national security concerns (confirmed, Wikipedia) — targets company/hardware supply chain, not model public access. - 2025-01-20: Trump signs EO delaying TikTok/PAFACA enforcement, establishing precedent for executive discretion on foreign-adversary app bans (confirmed). - 2025-02-07: HR.1121 "No DeepSeek on Government Devices Act" introduced (Gottheimer); scoped to federal IT only (confirmed, congress.gov). - 2025-02-27: S.765 companion bill introduced (Rosen); referred to committee, no further action since (confirmed). - 2025 (various): NY, TX, VA and agencies (Navy, NASA) block DeepSeek on devices/networks; reported as patchwork, not federal public ban (reported, MSSPalert/FedScoop). - 2026-01: RASA (Remote Access Security Act) passes House, expanding export controls to cloud-based chip access; awaiting Senate action (reported, CNBC). - 2026-07-14/07-15: Unrelated Senate calendar bills (Protecting American Taxpayers Act, WRDA 2026) advance — no AI content (confirmed, congress.gov). - 2026-07-22/27: NDAA FY2027 (HR8800, S.4784) advances in House/Senate — potential vehicle for China-tech provisions, no confirmed DeepSeek/public-access language (reported). - 2026-07-23: White House official confirms Moonshot AI accessed Nvidia chips despite export ban (reported, CNBC). - 2026-07-28: Trump admin moves to ban new Chinese robots/inverters — expanding China-tech restrictions but not AI models (reported, CNBC). - 2026-07-31: US threatens sanctions against Chinese AI labs (DeepSeek, Moonshot, MiniMax) over model distillation/IP theft (reported, multiple outlets). - 2026-08-01: Reports say China weighing own limits on foreign AI access as US "weighs restrictions" — signals escalation but no formal US action confirmed (reported, Breitbart/AsiaOne). - 2026-08-05/08-11: Intelligence Authorization Act FY2027 filed; FCC reported as "spearheading" Trump China tech policy (reported). - 2026-09 (by): DeepSeek prohibited across all Commerce Dept bureaus (device-level directive) — most recent confirmed federal action, still not public-facing (confirmed, MSSPalert). # Event Will the US federal government take formal action in 2026 that generally removes US public access to a major Chinese AI model (DeepSeek, ERNIE, Alibaba/Qwen, ByteDance, Moonshot, MiniMax, Hunyuan, or Z.ai/GLM)? # Outcomes to forecast Yes / No # Kalshi market anchor No direct Kalshi YES price was returned by kalshi_direct in this research pull (data gap). Related Kalshi markets (China EUV access, China GDP overtake, US-China FTA) show general China-restriction pricing in the 11–18% range, but none are close substitutes. Polymarket is used as the primary quantitative anchor instead (see below). # Sub-question answers 1. **Polymarket price/trend** — Currently 20.5% YES; up +7pts (7d) and +8pts (30d); range 7.5%–28.5% over 67 days; volume modest ($44k). (Polymarket direct) 2. **Congressional bills** — HR.1121 and S.765 ("No DeepSeek on Government Devices Act") remain stuck in committee since Feb 2025 with no floor action through mid-2026; scope limited to federal devices, not public access. Other bills (DEEP Act, Blocking CCP Spy Tech Act, NDAA FY2027, Intel Authorization FY2027) touch China/AI security broadly but no bill text confirmed to ban public consumer access. (congress.gov, claude_news) 3. **BIS/FCC/White House rulemaking** — No finalized ICTS/BIS rule targeting Chinese AI models for consumers found (only 2024 Kaspersky ICTS precedent exists). FCC reported as an active China-policy actor (Aug 2026) but no specific AI-model rule confirmed. White House reportedly "weighed" app-store bans/cloud restrictions in 2025 but these remain unconfirmed/early-stage as of Sept 2026. (claude_news, gdelt) 4. **US-China trade/tech stance** — Escalating: sanctions threats over AI distillation (Jul 2026), new bans on Chinese robots/inverters (Jul 2026), Moonshot Nvidia-chip export-ban circumvention revealed (Jul 2026), FCC-led aggressive China policy (Aug 2026). TikTok precedent (PAFACA) shows executive willingness to delay/soften bans despite law on books. (gdelt, wikipedia) 5. **State/agency bans & momentum** — NY, TX, VA and Navy/NASA have device-level bans; Commerce Dept-wide device ban confirmed by Sept 2026. No momentum evidence for a TikTok-style divest-or-ban law targeting DeepSeek's public availability. (claude_news) 6. **Timeline feasibility** — Given HR1121/S765 stalled 18+ months, and no BIS rule yet proposed, a full rulemaking/legislative process (NPRM, comment period, finalization) started now would be tight to complete and take effect by Dec 31, 2026, though an EO or ICTS "Final Determination" could move faster than legislation (Kaspersky precedent ~1 year). (inferred) 7. **Related markets signal** — No direct DeepSeek/Chinese-AI-ban Polymarket markets found. Kalshi related markets (China EUV, China GDP, US-China FTA) imply China-restriction base rates of ~11–18%, roughly consistent with sub-25% pricing for this event. (polymarket_related, kalshi_related) # Key facts (high-confidence, factual) 1. [congress.gov] HR.1121/S.765 limited to government devices; no public-access ban proposed in text. 2. [MSSPalert] Commerce Dept-wide DeepSeek device ban confirmed by Sept 2026 — govt-only. 3. [Wikipedia] TikTok/PAFACA precedent shows a real public-facing ban law exists but enforcement was repeatedly delayed by EO. 4. [CNBC] RASA passed House (Jan 2026), targets remote chip/cloud access, not model app bans; stuck in Senate. 5. [gdelt/multiple] US threatening sanctions on Chinese AI labs over distillation/IP (Jul 2026) — rhetoric, not enacted action. # Cross-market signals - Kalshi related: China EUV (11%), China GDP overtake (18%), US-China FTA (15%) — general China-restriction/relations base rates, not model-specific. - Polymarket: This event trading 20.5%, uptrend +8pts/30d suggesting rising but still minority expectation. - Sportsbook implied: N/A. # Analyst opinions and speculation - Claude_news synthesis: "no confirmed 2026 rule finalizing an app-store-wide or public ban," bottom line leans No given current trajectory. - Reporting suggests app-store ban/cloud restriction ideas remain in "early stages" since 2025 with no 2026 finalization confirmed. - Rising distillation/IP-theft rhetoric (Jul-Aug 2026) could accelerate action but currently targets sanctions/entity-list, not consumer app removal. # Directional lean per outcome - **Yes**: Escalating rhetoric (sanctions threats, robot/inverter bans, FCC activism), NDAA/Intel Authorization vehicles pending, TikTok precedent shows public bans are legally and politically feasible; Polymarket trending up. - **No**: 18+ months of stalled device-only legislation, no BIS/ICTS NPRM issued, all actions so far are government/device-scoped, TikTok precedent also shows EOs can delay/soften bans rather than trigger them, short remaining timeline (through Dec 2026) for full rulemaking. # Gaps / unknowns - No Kalshi direct YES price was retrieved for this ticker — must rely on Polymarket (20.5%) as proxy anchor. - No confirmed BIS/ICTS proposed rule text found; unclear if any is in the pipeline. - Unclear whether NDAA FY2027 or Intel Authorization FY2027 final text contains AI-model provisions. - No polling on Chinese AI ban among special-interest/insider forecasters found. # Calibration anchors - Polymarket YES: 20.5% (up from ~12.5% a month ago) — primary quantitative anchor given missing Kalshi direct data. - Precedent: TikTok ban legislated in 2024 but enforcement delayed repeatedly by EO — shows gap between "law on books" and "actual removal of access," relevant to this market's strict "must take effect" resolution criterion. - Precedent: 2024 Kaspersky ICTS Final Determination took ~1 year from investigation to enforcement — comparable timeline would need action to start almost immediately to bind by Dec 31, 2026.

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

claude-fable-5-1

55% conf

Yes	15%
No	85%

No Kalshi direct price was available, so the primary anchor is Polymarket at 20.5% YES (trending up from ~12.5%). The brief shows all confirmed federal actions through Sept 2026 are device/network-level bans on government systems (Commerce Dept, Navy, NASA) and the only relevant bills (HR1121/S765) are scoped to government IT and stalled 18+ months; no BIS/ICTS NPRM or app-store rule has surfaced. With roughly a quarter left in 2026 and the Kaspersky ICTS precedent taking ~1 year, a public-access removal that actually takes effect by Dec 31 would require an unusually fast EO or emergency determination, which the TikTok precedent (repeated enforcement delays) suggests is not the administration's default. Escalating rhetoric (distillation sanctions threats, FCC activism, NDAA vehicles) keeps a meaningful tail, so I shade only modestly below the Polymarket price.

gpt-5.6-sol

69% conf

Yes	17%
No	83%

With no direct Kalshi price available, the primary anchor is Polymarket's 20.5% YES price, supported loosely by related Kalshi signals in the 11–18% range. The legislative evidence points toward No because existing bills have been stalled for 18 months and apply only to government devices, while no consumer-facing BIS or ICTS rule has been proposed. Escalating sanctions rhetoric, FCC activism, and pending defense legislation preserve a fast-action tail risk, but no confirmed measure currently targets general public access. The short remaining 2026 timeline and the TikTok and Kaspersky precedents justify moving YES modestly below the Polymarket anchor.

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

Both forecasts converge tightly (0.15 vs 0.17) around the Polymarket anchor without independently stress-testing whether Polymarket's own +8pt/30-day uptrend and thin $44k volume make it a reliable anchor — a low-liquidity market with a strong recent trend could be noise-driven or could reflect real insider signal about NDAA/Intel Authorization text neither forecaster could verify, and this uncertainty isn't reflected in wider intervals.

Neither forecast explicitly addresses the resolution-criteria ambiguity flagged in the brief itself: the event asks about action that "generally removes public access," but the TikTok/PAFACA precedent shows a law can pass and still not result in actual removal due to EO delays — both forecasters treat "law enacted" and "access removed" as roughly equivalent when discounting, but should clarify whether a passed-but-unenforced ban (e.g., NDAA rider not yet implemented by Dec 31) would count as Yes or No, which materially affects calibration.

Both cite the NDAA FY2027 and Intelligence Authorization FY2027 as "pending vehicles" but neither investigates the base rate for controversial add-on provisions surviving conference/reconciliation by year-end — NDAA riders on tech bans historically have mixed survival rates, and this sub-question is waved off as unconfirmed rather than probability-weighted.

The escalation trajectory (sanctions threats on DeepSeek/Moonshot/MiniMax, robot/inverter bans, Moonshot Nvidia circumvention revelation, FCC activism) is uniformly described by both as "rhetoric" building a "tail" — but the brief shows this all occurred within a single month (Jul-Aug 2026), suggesting accelerating momentum that neither forecaster models as compounding risk into Q4 2026, potentially underweighting a scenario where distillation/IP-theft sanctions morph into an ICTS Final Determination faster than the Kaspersky 1-year baseline (e.g., emergency/IEEPA authority bypassing normal NPRM timelines).

Neither forecaster discusses the multi-model structure of the question (DeepSeek, ERNIE, Qwen, ByteDance, Moonshot, MiniMax, Hunyuan, Z.ai/GLM) — since Yes triggers if ANY major model's public access is removed, and Zhipu AI/Z.ai already faces Entity List blacklisting (Jan 2025) plus Moonshot/MiniMax face fresh sanctions threats, the disjunctive "any of 8 models" framing arguably supports a higher probability than a single-model-focused analysis (as done here, centered mostly on DeepSeek) would suggest.

Both forecasts lack any explicit tail-scenario for unilateral executive action via IEEPA or a national-emergency declaration (distinct from the slower BIS/ICTS/legislative path), which the brief's TikTok/PAFACA discussion implies is a live executive tool that could move faster than either forecaster's "rulemaking takes ~1 year" framing assumes.

6. Aggregator Fable 5.1 · final calibrated forecast

18% (62% confidence)

Yes: 18% No: 82%

The Polymarket anchor (20.5%, trending up) is the best quantitative signal given no Kalshi price, and both forecasters shaded modestly below it for sound reasons: all confirmed federal actions remain government-device bans, the relevant bills have been stalled 18+ months, no BIS/ICTS NPRM exists, and the strict 'must take effect' criterion combined with only ~3.5 months left and the TikTok delay precedent makes actual public removal hard. However, the critique correctly notes the disjunctive structure (any of 8 models), the live OFAC/IEEPA fast-path risk implied by July's sanctions threats against DeepSeek/Moonshot/MiniMax (an SDN designation could functionally remove app-store access without rulemaking), and that a thin market's sustained uptrend may reflect real momentum rather than noise. Balancing these, I stay close to the market rather than discounting as far as the forecasters did, landing at 18% Yes.

Pipeline Timing

Total pipeline time: 246.5s

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