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

date	predicted	market price	confidence
2026-09-09	18%	20%	62%

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 price and recent trend for this market, and are there related Polymarket/Kalshi markets on DeepSeek/Chinese AI bans?
Which bills in the 119th Congress would ban or restrict Chinese AI models (e.g., Decoupling America's AI Capabilities from China Act, No DeepSeek on Government Devices Act), and how far have they advanced beyond government-device-only restrictions?
Has the Commerce Department (BIS/ICTS) or the White House initiated any rulemaking, entity listing, or ICTS supply-chain action (like the TikTok/Kaspersky precedents) that could remove public access to a Chinese AI model within the US?
Is there evidence the Trump administration favors a general consumer ban versus narrower measures (federal device bans, export controls on chips), and what have officials like Lutnick, Sacks, or Rubio said about DeepSeek/Qwen availability in the US?
What is the historical base rate and timeline for US actions that actually removed public access to a Chinese app/software (TikTok, Huawei, Kaspersky, DJI), and how long did they take from proposal to effective removal?
Have any Chinese AI companies (DeepSeek, Alibaba, ByteDance, Moonshot, MiniMax, Zhipu) been added to the Entity List or otherwise targeted in 2025-2026, and would such actions actually remove app-store/API/open-weight access for US users?
Are Chinese models currently accessible to US users through many channels (app stores, Hugging Face open weights, US cloud hosting), making 'general removal' technically and legally hard to achieve?

Planner reasoning

This is a Polymarket question about whether the US federal government will take formal action removing general US public access to a flagship Chinese AI model (DeepSeek, Qwen, ERNIE, etc.) by end of 2026. Key drivers are pending legislation (e.g., 'No DeepSeek on Government Devices Act', 'Decoupling America's AI Capabilities from China Act'), Commerce/BIS export-control or ICTS actions, executive orders, and whether any action would go beyond government devices to a general public ban. Base rate: TikTok-style bans took years and still didn't fully remove access, so the polymarket price plus news on legislative/regulatory momentum are the main anchors.

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): 16.50% - 7-day price change: -4.00% - 30-day price change: +3.00% - Total volume: $51,720 (USD notional) - Price range: 7.50% - 28.50% - Data points: 74 days
polymarket_related	OK	3.2s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'DeepSeek': 0 markets \| keyword 'Chinese AI': 0 markets \| keyword 'AI ban': 0 markets \| keyword 'Qwen': 0 markets
kalshi_related	OK	3.0s	2	2 related markets / summaries. keyword 'DeepSeek': no matches \| keyword 'Chinese AI ban': ok \| keyword 'China AI': ok
congress_bills	OK	1.3s	10	Found 10 bills matching 'DeepSeek'.
congress_bills	OK	1.0s	6	Found 6 bills matching 'Chinese artificial intelligence ban'.
article_search	OK	0.8s	38	Fetched 38 unique articles across 5 queries (lookback=90d). 'US ban DeepSeek public access': 10 articles \| 'Commerce Department Chinese AI model restriction ICTS': 10 articles \| 'executive order Chinese AI models United States': 10 articles \| "Decoupling America's AI Capabilities from China Act": 10
claude_news	OK	32.3s	13	Based on my research, here are the key findings: - Government-devices-only bans already in effect: The Commerce Department and Navy banned DeepSeek on government-furnished equipment in early 2025, with one Commerce email stating "access to the new Chinese based AI DeepSeek is broadly prohibite
gdelt_news	OK	208.9s	10	GDELT: 10 articles across 3 queries (lookback=60d). 'DeepSeek ban United States': error GDELT rate-limited after retries (429) \| 'Chinese AI model export control': error GDELT rate-limited after retries (429) \| 'Commerce Department DeepSeek': 10 hits
wikipedia	OK	0.1s	3	Fetched 3 Wikipedia entries (0 missing pages).

3. Evidence Brief Sonnet 5 · 9037 chars

# Current state As of late 2026 (market close 2026-12-31), no US federal action has removed general public access to any major Chinese AI model (DeepSeek, Qwen, ERNIE, Hunyuan, Kimi, etc.). Current measures are limited to government-device/procurement bans (already in effect at some agencies since early 2025) and pending legislation; internal Trump administration debate over broader restrictions is ongoing but unresolved, with pro-market voices (Sacks) resisting outright bans favored by security hawks. # Timeline of key events - 2025-01-20 to 2026-01-22: TikTok (PAFACA) saga — precedent for how long a "ban" takes to become genuinely effective (de jure ban Jan 2025, repeatedly delayed via EO, resolved via divestiture Jan 2026) — confirmed (Wikipedia). - 2025-02 (early 2025): Commerce Dept divisions and Navy ban DeepSeek on government-furnished devices — confirmed (Engadget). - 2025-02-07/27: H.R.1121 and S.765 "No DeepSeek on Government Devices Act" introduced, referred to committee — confirmed, no further advancement (Congress.gov). - 2026-01 (~Jan 2026): 21 state AGs urge Congress to pass DeepSeek government-device ban; note other countries (Canada, Australia, South Korea, Taiwan, Italy) went further — reported (Alabama AG office). - 2026-07-20: Axios reports Trump administration reviving push against Chinese open-weight models post-Kimi K3 launch (NSA/ONCD advisory, draft EO on liability for hosting Chinese models); paused earlier over market-impact concerns — reported (Axios). - 2026-07-20: David Sacks publicly opposes a ban, framing it as regulatory capture by closed-lab incumbents — confirmed statement (Axios/X). - 2026-07-23: CNBC reports Moonshot AI accessed Nvidia chips despite export ban; RASA (Remote Access Security Act) passed House Jan 2026, pending Senate — reported. - Mid-2026: Departure of pro-market adviser Sriram Krishnan reported to renew momentum for restrictions within administration — reported (BigGo Finance). - 2026-09-01: SCMP reports Congress advancing a bill to boost US open-source AI competitiveness vs. China (not a ban) — reported. - 2026-09-12: Reports on Chinese labs' large-scale distillation of Claude outputs, fueling security/IP arguments for restriction — reported (Tech Times). - Through 2026-09: No Entity List addition, executive order, or export control has actually removed consumer/public access to any major Chinese AI model in the US. # Event Will the US federal government take formal action (legislation, EO, export control, or other) that generally removes US public access to a major Chinese AI model (DeepSeek, ERNIE, Qwen, Hunyuan, MiniMax, Kimi, Zhipu) between market creation and Dec 31, 2026? # Outcomes to forecast - Yes - No # Kalshi market anchor No direct Kalshi order-book price was returned; closest proxy is Polymarket's identical-ticker market at **16.5% YES** (current), down 4pts over 7 days, up 3pts over 30 days, range 7.5%-28.5% over 74 days, modest volume (~$52K total). No matching Kalshi-native market found (keyword "DeepSeek" returned no hits); related Kalshi China-policy markets (EUV, GDP overtake, FTA) trade 11-15%, suggesting general skepticism toward major US-China policy actions completing within timeframe. # Sub-question answers 1. **Polymarket price/trend & related markets** — Current 16.5%, 7d -4%, 30d +3%; no other DeepSeek/Chinese AI/AI ban/Qwen Polymarket markets found (Polymarket direct/related tools). 2. **Congressional bills** — S.765/H.R.1121 "No DeepSeek on Government Devices Act" (Feb 2025) remain in committee, unadvanced; these and similar bills (Blocking CCP Spy Tech Act, DEEP Act) target only government devices/procurement, not general public access; no broader consumer-ban bill has passed either chamber (Congress.gov, claude_news). 3. **Commerce/BIS/ICTS rulemaking** — No confirmed BIS Entity List addition or ICTS rule targeting Chinese AI model developers found; only a draft/considered EO (liability for hosting Chinese models) and NSA/ONCD advisory reported as under discussion, not issued (Axios, BigGo). 4. **Administration stance (Lutnick/Sacks/Rubio)** — David Sacks explicitly opposes a ban, calling it regulatory capture (Axios, 7/20/26). No direct Lutnick/Rubio quotes on DeepSeek/Qwen found in research. Reported administration preference is for "soft" measures (advisories, procurement pressure, boosting US open-source) rather than outright consumer bans, though momentum for stricter action reportedly grew after Krishnan's departure (BigGo). 5. **Historical base rate/timeline (TikTok, Huawei, Kaspersky, DJI)** — TikTok took ~5 years (2020 proposal to 2025 nominal ban) and even then the "ban" was repeatedly delayed via EO non-enforcement, resolved via divestiture rather than access removal (Wikipedia). This suggests long lag and high likelihood of dilution/non-enforcement even after formal action. 6. **Entity List/targeting of Chinese AI firms in 2025-26** — No confirmed Entity List additions for DeepSeek, Alibaba, ByteDance, Moonshot, MiniMax, or Zhipu found in research; "threat of Entity List" is discussed as leverage, not enacted (BigGo Finance, claude_news). 7. **Technical/legal feasibility of "general removal"** — Chinese models are broadly available via app stores, Hugging Face open weights, and US cloud hosting (Azure, AWS host some); analysts (Tom's Hardware) note open-weight downloadability makes an outright ban "nearly impossible to enforce," reducing likelihood of a clean qualifying event even if formal action is taken. # Key facts (high-confidence, factual) 1. [Engadget] Commerce Dept/Navy banned DeepSeek on government devices, early 2025 — device-only, not public-access removal. 2. [Congress.gov] S.765/H.R.1121 remain in committee as of research date; not enacted. 3. [Wikipedia] TikTok ban precedent: 5-year gap from 2020 proposal to 2025 nominal enforcement, then reversed/delayed repeatedly, resolved via divestiture not access removal. 4. [Axios] Trump administration considered but paused broader restrictions (NSA/ONCD advisory, EO on hosting liability) as of July 2026; not issued. 5. [Polymarket] Market itself prices this at 16.5%, well below 50%, with declining 7-day momentum. # Cross-market signals - Kalshi related: No direct China-AI-ban market; adjacent China-policy markets (EUV 11%, GDP overtake 14%, FTA 15%) all price big US-China structural actions in the 11-15% range — a possible reference class anchor. - Polymarket: Same-ticker market at 16.5%, down from 28.5% high, suggesting fading probability as time passes without escalation. - Sportsbook implied: N/A (not applicable to this event type). # Analyst opinions and speculation - Axios/Tom's Hardware: open-weight distribution makes true "public access removal" technically near-impossible to enforce even if a ban is issued. - BigGo Finance: departure of pro-market advisers (Krishnan) seen as strengthening restriction faction internally — a bullish-for-Yes signal but unconfirmed as policy outcome. - Sacks (public figure, not neutral analyst): actively lobbying against any ban, indicating contested administration consensus. - Just Security (Aug 2026): proposes "test, standardize, restrict" framework — indicates policy community still in proposal stage, not implementation. # Directional lean per outcome - **Yes**: Supported by active internal deliberation (draft EO, NSA/ONCD advisory), rising bipartisan legislative activity, RASA passing House, and precedent of Trump admin acting suddenly (TikTok EOs) when it chooses. Opposed by: no formal action taken yet in ~20 months since DeepSeek's Jan 2025 emergence, Sacks/open-source lobby pushback, technical infeasibility of true "general removal" given open weights, and only ~4 months remaining in window. - **No**: Supported by Polymarket pricing (16.5%), historical slowness of Chinese-app bans (TikTok took years and was diluted), administration's stated preference for advisories/procurement over consumer bans, and lack of any Entity List action to date. Opposed by: possibility of a sudden EO (per TikTok precedent) or NDAA rider (S.4784/HR.8800 FY2027 NDAA in progress, could carry a provision). # Gaps / unknowns - No visibility into final text/status of FY2027 NDAA (S.4784/HR.8800) — could contain a Chinese-AI provision; unresolved as of latest data (2026-09-14, "received in Senate"). - Unclear whether any single-channel removal (e.g., app store delisting without full access removal) would sway resolution — rules require removal "across ordinary public channels," a high bar. - No confirmed direct quotes from Lutnick/Rubio on this specific issue. # Calibration anchors - Polymarket current YES price: 16.5% (primary anchor), down from 30-day high of 28.5%. - Adjacent Kalshi China-policy markets cluster at 11-15%, reinforcing skepticism of major bilateral US-China action within calendar-year windows. - TikTok precedent: even successful "bans" took years and were substantially delayed/diluted before effect — supports base rate below 20% for full effect within ~1 remaining year.

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

claude-fable-5-1

55% conf

Yes	12%
No	88%

The closest market anchor (Polymarket, identical ticker) sits at 16.5% YES and has been drifting down as the window shrinks with no formal action taken; adjacent Kalshi China-policy markets cluster at 11-15%. In ~20 months since DeepSeek's emergence, federal action has been limited to government-device bans and stalled committee bills, while the administration's own draft EO (hosting liability) and NSA/ONCD advisory were paused and face open opposition from Sacks and the open-source lobby. The resolution bar—removal of general public access 'across ordinary public channels'—is very high given open-weight distribution via Hugging Face and US cloud hosting, meaning even a sudden EO or NDAA rider would likely not cleanly qualify by Dec 31. The TikTok precedent (years of delay and dilution) further supports a below-market probability, so I shade modestly under the 16.5% anchor to reflect the short remaining window and stringent resolution criteria.

gpt-5.6-sol

74% conf

Yes	13%
No	87%

The 16.5% Polymarket YES price is the primary anchor, while adjacent Kalshi China-policy markets at 11-15% also imply skepticism. Congressional proposals remain confined to government devices or procurement, and no broader bill, Entity List action, ICTS rule, or executive order has removed public access. Although reported draft measures and renewed security concerns create some chance of sudden action, administration resistance and the preference for softer measures point toward No. I move modestly below the market anchor because only a few months remain, open-weight distribution makes qualifying removal difficult, and the TikTok precedent shows that even determined restrictions tend to take years or be diluted.

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

Both forecasters converge tightly (12% and 13%) and essentially just shade slightly below the 16.5% Polymarket anchor without independently stress-testing the FY2027 NDAA wildcard (S.4784/HR.8800), which the brief explicitly flags as an unresolved gap that "could contain a Chinese-AI provision" — neither rationale mentions NDAA risk at all, a notable omission given it's the most concrete near-term legislative vehicle.

Neither forecast grapples with the ambiguity in resolution criteria noted in the "Gaps/unknowns" section — that a partial or single-channel removal (e.g., cloud-hosting liability EO effectively forcing Azure/AWS to drop Chinese models, or app-store delisting) might or might not count as "general public access removal." This ambiguity could push probability either direction depending on strict vs. loose interpretation, and both forecasters treat the bar as uniformly "very high" without acknowledging the interpretive risk that markets/resolvers might apply a looser standard.

Both anchor heavily on Polymarket's 16.5% and shade only slightly downward (to ~12-13%), but neither seriously engages with the countervailing bullish signals: RASA passing the House, Krishnan's departure "renewing momentum for restrictions," and the post-Kimi K3 NSA/ONCD advisory — these are treated as boilerplate mentions rather than actively weighed against the base-rate anchor, risking underweighting a plausible acceleration scenario in Q4 2026.

The TikTok analogy is invoked by both forecasters as supporting a low probability, but this may be a flawed reference class: TikTok involved a single dominant app with a very visible divestiture negotiation, whereas Chinese open-weight LLMs (DeepSeek, Qwen, Kimi) have no clear "divestiture" resolution path and multiple companies/models, meaning if the administration acts, it may take a blunter and faster EO/export-control route rather than the years-long TikTok process — this reference-class risk is underexplored.

Neither forecaster discusses the confirmed fact that "Moonshot AI accessed Nvidia chips despite export ban" (CNBC) and the growing distillation/IP controversy (Claude-output distillation reports, Sept 2026) as potential catalysts that could trigger rapid, symbolic enforcement action in Q4 2026 distinct from the slow-moving legislative track — this is a plausible near-term trigger neither model incorporates.

Both forecasts express very similar confidence levels in their point estimates (0.55 and 0.74) despite nearly identical numeric outputs, suggesting the stated confidence isn't clearly tied to differentiated reasoning; given the genuine uncertainty flagged in "Gaps/unknowns" (NDAA status, resolution ambiguity, no Lutnick/Rubio quotes), a more explicit acknowledgment of wide uncertainty bands (e.g., 8-20%) rather than tight convergence around 12-13% might better reflect the evidence's actual dispersion.

6. Aggregator Fable 5.1 · final calibrated forecast

14% (64% confidence)

Yes: 14% No: 86%

Both forecasters and the Polymarket anchor (16.5%, drifting down) agree that no federal action has removed general public access to any major Chinese AI model, and the only concrete measures remain government-device bans and stalled committee bills. The critique correctly notes the forecasters ignored the FY2027 NDAA vehicle and resolution ambiguity around a hosting-liability EO, which justifies not shading as far below market as they did, but NDAA provisions historically target procurement rather than consumer access, and the draft EO remains paused amid open opposition from Sacks and the open-source lobby. With only a few months left and a high bar ('across ordinary public channels') that open-weight distribution makes hard to meet, time decay argues for staying modestly below the market price. I settle at 14% YES, between the forecasters' 12-13% and the 16.5% anchor.

Pipeline Timing

Total pipeline time: 310.8s

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