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

date	predicted	market price	confidence
2026-08-26	13%	18%	55%
2026-08-19	19%	22%	56%

1. Planner Fable 5.1 · Fermi decomposition + tool plan

Tools chosen: polymarket_direct polymarket_related kalshi_related article_search claude_news congress_bills congress_bills gdelt_news

Sub-questions (Fermi decomposition)

Which major AI model was already removed from US public access by formal US government action in 2025-2026, what legal mechanism was used (executive order, Commerce ICTS rule, export control, legislation), and how long did it take from announcement to effect?
What pending or proposed US federal actions (Commerce Department ICTS rulemakings, executive orders, FCC/Treasury actions) specifically target Chinese AI models from DeepSeek, Alibaba, ByteDance, Moonshot AI, or Zhipu AI, and what are their expected effective dates?
What is the legislative status of bills that would ban Chinese AI models in the US (e.g., Decoupling America's AI Capabilities from China Act, No DeepSeek on Government Devices Act, Chip Security / Remote Access Security Acts), and do any extend to general public access rather than just government devices?
Are the remaining listed Chinese models (Qwen, Doubao/Seed, Kimi, GLM) currently accessible to the US public via ordinary channels (app stores, web, API), such that a ban would constitute a 'general removal' under the resolution criteria?
Is there any plausible US government action against a domestic major model (xAI Grok, Meta Llama, OpenAI, etc.) — e.g., safety-related enforcement, court-ordered takedown, or export control — that could remove general public access in 2026?
What is the current Polymarket YES price and trend, and do related Polymarket/Kalshi markets (DeepSeek ban, Chinese AI ban, TikTok-style actions) imply a consistent probability?

Planner reasoning

The title's 'another' implies a prior formal US removal of a major AI model (likely a Chinese model such as DeepSeek) has already occurred by market creation in July 2026, so the key questions are what mechanism was used, whether similar actions (Commerce ICTS rules, executive orders, legislation like the Decoupling America's AI Capabilities from China Act) are pending against other listed models (Alibaba Qwen, ByteDance, Moonshot Kimi, Zhipu GLM), and how quickly they could take effect before Dec 31, 2026. The Polymarket price is the primary anchor; news, congressional bill status, and cross-market signals 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 another major AI model in 2026? - Current price (probability): 20.00% - 7-day price change: +10.50% - 30-day price change: -1.50% - Total volume: $26,159 (USD notional) - Price range: 8.50% - 45.00% - Data points: 75 days
polymarket_related	OK	7.0s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'DeepSeek ban': 0 markets \| keyword 'AI model ban': 0 markets \| keyword 'Chinese AI': 0 markets \| keyword 'Qwen': 0 markets \| keyword 'executive order AI': 0 markets
kalshi_related	OK	6.8s	2	2 related markets / summaries. keyword 'DeepSeek': no matches \| keyword 'AI model ban': ok \| keyword 'Chinese AI ban': ok
article_search	OK	0.8s	40	Fetched 40 unique articles across 5 queries (lookback=120d). 'US government bans DeepSeek public access': 10 articles \| 'Commerce Department ICTS rule Chinese AI models': 10 articles \| 'executive order removing access AI model United States': 10 articles \| 'Alibaba Qwen ByteDance Moonshot Zhipu US b
claude_news	OK	29.7s	16	## Findings Already-occurred removal (precedent event in 2026): - On June 12, 2026, Commerce (BIS) sent Anthropic an "Is-Informed Letter" export-control directive under ECRA requiring a license before any foreign national—anywhere, including Anthropic's own foreign employees—could access its ne
congress_bills	OK	1.0s	5	Found 5 bills matching 'Chinese artificial intelligence ban DeepSeek'.
congress_bills	OK	0.9s	3	Found 3 bills matching "Decoupling America's Artificial Intelligence Capabilities from China".
gdelt_news	OK	76.4s	30	GDELT: 30 articles across 3 queries (lookback=60d). 'US ban Chinese AI models Qwen Kimi': 10 hits \| 'Commerce Department AI model prohibition': 10 hits \| 'DeepSeek removed app store US government': 10 hits

3. Evidence Brief Sonnet 5 · 7459 chars

# Current state One qualifying-style action already occurred in 2026 (Commerce/BIS export-control letter forced Anthropic to pull Claude Mythos 5 and Fable 5 from general US access) but was **reversed within ~3 weeks**, restoring access by late June/early July. This market asks whether **another, distinct** major model has its US public access formally removed (for any duration) before Dec 31, 2026. As of the latest research (mid-to-late Aug 2026), no second removal has occurred; several mechanisms targeting Chinese models (Entity List, ICTS rule, AI Diffusion Framework rewrite) are under active consideration but not yet enacted, while DeepSeek/Qwen/Kimi/GLM remain broadly accessible. # Timeline of key events - 2026-06-12 (confirmed): Commerce/BIS sends Anthropic an "Is-Informed Letter" requiring an export license for any foreign national access to Claude Mythos 5/Fable 5; Anthropic, unable to screen users technically, pulls both models from all US public access. - 2026-06-26 (confirmed): Mythos 5 access restored for some approved US organizations. - 2026-06-30/07-01 (confirmed): Commerce fully lifts export controls; Fable 5 restored to general access. - 2026-07-17 (reported): CNBC — White House informally "gating" frontier releases (voluntary 30-day pre-release review executive order); no legal mandate. - 2026-07-20 (reported): Axios — Trump administration weighing Entity List designation for Chinese AI labs (esp. Moonshot/Kimi) over alleged IP theft/distillation; not yet implemented. - 2026-07-23 (reported): CNBC/Cryptonomist — Treasury Sec. Bessent warns of possible sanctions/Entity List action against Moonshot; Remote Access Security Act (RASA) passed House (Jan 2026), pending in Senate. - 2026-08-15/16 (confirmed): Alibaba's Qwen models surpass 3B downloads, overtaking Meta/Google in open-source usage — indicates no current restriction on Chinese open models. - Ongoing/expected by 2026-09-30: Commerce's "Redesigned AI Diffusion Framework" rule (withdrawn March 2026, recommitted for FY2026) and a "remote-server loophole" export rule reportedly could be shared with industry as early as September 2026 (reported, not enacted). # Event Will the US federal government take formal action (law, EO, export control, etc.) that generally removes US public access to a second major AI model (beyond the already-reversed Anthropic case) at any point in 2026? # Outcomes to forecast Yes / No # Kalshi market anchor No direct kalshi_direct price was returned in research. The identical-ticker market on Polymarket (0x39ce5b38…) shows: **current YES 20%**, 7-day change **+10.5%**, 30-day change **-1.5%**, range 8.5%–45% over 75 days, volume ~$26k. This should be treated as the best available proxy anchor; recent upward momentum coincides with Moonshot/Entity List reporting (late July 2026). # Sub-question answers 1. **Prior removal & mechanism**: Anthropic's Claude Mythos 5/Fable 5 were removed from general US access via a Commerce/BIS export-control "Is-Informed Letter" (ECRA authority) on 2026-06-12, effective almost immediately; restored ~2–3 weeks later (claude_news). 2. **Pending Chinese-model actions**: No enacted rule yet. Under consideration: Entity List designation for Moonshot AI/Kimi (Treasury/Commerce, reported July 2026), ICTS rule (existing authority, unused on AI so far), and a "Redesigned AI Diffusion Framework" rule due by FY2026 end (Sept 30, 2026); a remote-server export-control rule possibly shared with industry as early as Sept 2026 (claude_news, Tom's Hardware, Axios — all reported, not confirmed enacted). 3. **Legislative status**: Decoupling America's AI Capabilities from China Act (S.321) stuck in Senate Judiciary Committee since Jan 2025, no further action. No DeepSeek on Government Devices Act covers federal devices only, not general public access. RASA passed House Jan 2026, pending Senate (congress_bills, claude_news). 4. **Current accessibility of Chinese models**: Qwen, Kimi, GLM, Doubao remain accessible via app stores/web/API; Qwen hit 3B downloads Aug 2026, surpassing Meta/Google (gdelt_news). DeepSeek also still generally accessible for consumer use (claude_news). 5. **Domestic model risk**: Anthropic precedent shows domestic models (via export-control mechanism) are plausible targets even without China nexus; White House also informally pressuring "voluntary" pre-release reviews (CNBC, 2026-07-17). No court-ordered takedown reported for OpenAI/Meta/xAI/Google. 6. **Cross-market consistency**: Polymarket YES=20%, trending up recently; no matching Kalshi/Polymarket sub-markets found for DeepSeek ban, Chinese AI ban, or Qwen specifically (polymarket_related, kalshi_related returned no relevant matches). # Key facts (high-confidence, factual) 1. [claude_news/CSIS/CNBC] Anthropic Mythos 5/Fable 5 were fully pulled from US public access via Commerce export-control letter (2026-06-12), restored by ~2026-07-01. 2. [congress_bills] Decoupling Act (S.321) has seen no movement since Jan 2025 referral. 3. [gdelt_news] Qwen models hit 3B downloads (Aug 2026), no US restriction evident. 4. [claude_news] RASA passed House Jan 2026; pending Senate. 5. [claude_news/Axios] Entity List/sanctions against Moonshot under "active consideration," not implemented as of late July 2026. # Cross-market signals - Kalshi related: No relevant matches found; searches returned unrelated markets (Fed presidency, transgender sports, stadium). - Polymarket (same-ticker market): 20% YES, +10.5% 7d, -1.5% 30d, volume thin (~$26k) — moderate confidence signal only. - Sportsbook implied: N/A. # Analyst opinions and speculation - PIIE: "ad hoc" export controls (Anthropic case) could backfire by pushing global users toward Chinese models — argues against government repeating the tactic casually. - Peter Harrell (X): outlines menu of options from "FUD" campaigns to formal ICTS/Entity List actions against Chinese AI — signals mechanism exists but political calculus undetermined. - Macron (G7): warned that arbitrary US model shutoffs damage credibility — mild disincentive against repeat action, especially against US firms. # Directional lean per outcome - **Yes**: Multiple pending mechanisms (Entity List, ICTS, AI Diffusion Framework rule due Sept 2026, remote-server rule) actively being developed for Chinese models; recent precedent (Anthropic) proves willingness/capacity to act fast; Polymarket trending up. - **No**: All Chinese models remain accessible with growing usage (Qwen 3B downloads); legislative bills stalled; no Entity List/rule formally issued as of Aug 2026; White House rhetoric emphasizes "voluntary" collaboration, not bans; only ~4 months remain post-August for a new formal action to both issue AND take effect. # Gaps / unknowns - No genuine kalshi_direct price retrieved for this exact ticker — anchor relies on same-ticker Polymarket data. - Unclear whether AI Diffusion Framework/remote-server rule, if issued, would cause a "general removal" (vs. narrower compute-access restriction) meeting resolution criteria. - No confirmation on Entity List decision timing or whether it would target public consumer access vs. enterprise/cloud access only. # Calibration anchors - Polymarket YES (same ticker): 20%, recently trending up from ~9-15% range. - Precedent: One qualifying-style event already occurred and reversed within 2026, establishing non-trivial base rate but showing reversals happen fast under industry/legal pushback.

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

claude-fable-5-1

42% conf

Yes	21%
No	79%

The best available anchor is the same-ticker Polymarket price at 20% YES, and the brief surfaces no enacted action beyond the already-reversed Anthropic Mythos 5/Fable 5 case. Pending mechanisms (Moonshot Entity List designation, ICTS rule, AI Diffusion Framework rewrite due ~Sept 30, remote-server export rule) show real intent, but an Entity List listing restricts exports *to* a firm and would not by itself remove US public access to Kimi, and the ICTS/Diffusion rules are unenacted with only ~4 months left for issuance plus effect. Counter-evidence is strong: Qwen at 3B downloads with no restriction, S.321 stalled since Jan 2025, RASA stuck in the Senate, and White House rhetoric emphasizing voluntary review rather than bans. The Anthropic precedent proves the administration can act fast via BIS letters, which keeps YES meaningfully above a naive base rate, so I land slightly above the market at 21%.

gpt-5.6-sol

59% conf

Yes	20%
No	80%

The Anthropic episode demonstrates that Commerce can remove access rapidly through export controls, modestly supporting Yes, but its quick reversal makes repetition less likely. Potential actions against Moonshot/Kimi and rules addressing AI diffusion or remote-server access provide plausible pathways, yet none had been enacted by mid-August and it is unclear whether they would remove general public access. Stalled legislation, continued broad availability of DeepSeek, Qwen, Kimi, and GLM, and only about four months remaining point toward No. With no direct Kalshi price, I retain the identical-ticker Polymarket consensus anchor of 20% Yes because the brief provides no strong evidence that the market has materially mispriced the pending-policy risk.

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

Both forecasters anchor tightly to the thin-volume ($26k) Polymarket proxy without adjusting for its momentum signal—+10.5% over 7 days coinciding with Entity List/Moonshot reporting suggests the market itself may be underpricing near-term catalysts (AI Diffusion Framework rule "due by Sept 30, 2026" and remote-server rule "shared with industry as early as Sept 2026"), which fall squarely within the resolution window and neither forecast explicitly weights this looming deadline.

Neither forecast adequately grapples with the ambiguity in resolution criteria noted in the brief's "Gaps/unknowns": it's unclear whether an Entity List action or the Diffusion Framework rule would constitute "general removal" of public access versus narrower enterprise/compute restrictions—this ambiguity could cut either way and deserves more explicit hedging in the probability rather than being mentioned only as a caveat.

Both forecasts underweight the base-rate lesson from the Anthropic precedent itself: a fast, unilateral BIS "Is-Informed Letter" required no legislative action and took effect "almost immediately" on 2026-06-12, meaning a second such letter (against any lab, not just Chinese ones per the "domestic model risk" sub-question) could materialize with little warning in the remaining four months, yet both forecasts treat legislative gridlock (S.321, RASA) as if it were the primary pathway.

The two forecasts are nearly identical (21% vs 20%) with little independent reasoning divergence, suggesting anchoring-induced convergence rather than genuine independent evidence synthesis—neither explores a scenario where the reversal pattern itself increases Yes probability (i.e., government appears willing to act fast now that legal/technical friction from the first case is resolved, lowering the cost of a repeat action).

Neither forecaster discusses the Macron/G7 credibility-cost argument or Anthropic's technical screening problem as a *structural* reason the mechanism might be reused specifically because it's low-cost/reversible (quick action, quick walk-back, minimal lasting diplomatic damage), which could argue for a higher Yes than the static "stalled legislation" framing suggests.

Both treat the "4 months remaining" constraint as strongly bearish, but the Anthropic case took only ~2 weeks from letter to enforcement, so the timeline objection is weaker than presented—an Entity List or ICTS action could plausibly be issued and take effect well within Q4 2026 if political will exists (as evidenced by Bessent's July 23 warning).

6. Aggregator Fable 5.1 · final calibrated forecast

23% (50% confidence)

Yes: 23% No: 77%

Both forecasters landed at ~20%, matching the same-ticker Polymarket proxy, and the brief confirms no second removal has occurred as of late August with all major Chinese models still broadly accessible. The critique's strongest point is that the Anthropic precedent shows a BIS 'Is-Informed Letter' can remove access within days without legislation, so the four-month window is less binding than the forecasts implied, and the AI Diffusion Framework and remote-server rules have reported September deadlines squarely inside the window. However, the quick reversal under industry and allied pushback raises the political cost of a repeat, an Entity List designation for Moonshot would restrict exports to the firm rather than necessarily remove US public access to Kimi, and it is unclear the pending rules would meet the 'general removal' bar. I nudge slightly above the thin-volume market to 23% Yes to reflect the demonstrated fast-action mechanism and near-term rule catalysts, while keeping No as the clear favorite.

Pipeline Timing

Total pipeline time: 213.9s

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