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

- [2026-06-02] President Trump signed EO "Promoting Advanced Artificial Intelligence Innovation and Security," creating a voluntary early-access framework for frontier models and explicitly stating it does not create "mandatory governmental licensing, pre-clearance, or permitting requirements" — it did not remove public access to any model. (https://www.whitehouse.gov/fact-sheets/2026/06/fact-sheet-president-donald-j-trump-promotes-advanced-artificial-intelligence-innovation-and-security/) - [2026-06-12] Commerce Department's Bureau of Industry and Security issued an export-control directive requiring Anthropic to deny Claude Fable 5 and Claude Mythos 5 to any foreign national; because Anthropic could not verify nationality in real time, it disabled both models for all US and global users — a precedent event, but it predates this market's July 3, 2026 creation date. (https://www.techpolicy.press/did-the-us-government-just-set-an-ai-export-precedent-by-blocking-mythos/) - [2026-06-30/07-01] The Commerce Department lifted the Fable 5/Mythos 5 export controls; Anthropic restored Fable 5 globally on July 1 and restored limited Mythos 5 access for US "Project Glasswing" partners, resolving the precedent episode before the market opened. (https://www.anthropic.com/news/redeploying-fable-5) - [2026-07-03] This Polymarket contract opened, explicitly framed as asking about "another" removal — i.e., the Fable/Mythos episode is treated as the baseline precedent, not a qualifying event for this market. (https://polymarket.com/event/us-government-removes-public-access-to-another-major-ai-model-in-2026-20260703202936862) - [2026-09-01] Anthropic released Claude Mythos 5.1 and Claude Fable 5.1 as successor models, with no further government restriction reported at launch. (https://en.wikipedia.org/wiki/Claude_Mythos) - [2026-05-26] A legal-status review found DeepSeek "is not banned nationwide in the United States for ordinary personal use," with restrictions limited to government, defense, intelligence and contractor contexts. (https://chat-deep.ai/privacy-security/is-deepseek-banned-in-us/) - [2026-05-26] FY2026 NDAA and intelligence-authorization provisions require removal of DeepSeek from DoD and covered intelligence-community systems, but do not reach the general public. (https://chat-deep.ai/privacy-security/is-deepseek-banned-in-us/) - [2026-09-18] House Select Committee on China Chairman John Moolenaar urged tighter export controls on "advanced AI products" aimed at China, but this concerns chips/products for China, not domestic public access removal. (https://exportcompliancedaily.com/topic/ai_export_controls) - [2026-10-02] A former State Department official wrote that the US "may be able to take some steps to impose license requirements" on dangerous uses of open-weight models, indicating this remains a live policy debate, not enacted action. (https://exportcompliancedaily.com/topic/ai_export_controls) - [2026-10-06] A law-review piece noted the Commerce Department's June Anthropic directive was "a nonpublic directive" and argued such export-control use for AI services needs public rulemaking — reflecting ongoing scrutiny of whether this tool will be used again, with no new instance yet. (https://www.theregreview.org/2026/10/06/oktenli-when-a-chatbot-becomes-an-export/) - [2026-10-02] Reporting on US-China trade/AI talks (Sept. 20) states export controls "weren't discussed" in the latest US-China AI dialogue, and chip-related export-control bills remain stalled in the Senate. (https://exportcompliancedaily.com/topic/ai_export_controls) BOTTOM LINE: The one clear precedent of a US government action causing general public removal of a major AI model (Anthropic's Fable 5/Mythos 5, June 12–30, 2026) occurred and was fully resolved before this market's July 3, 2026 creation, and no comparable action against any major model's general US public access has occurred since; current policy (June 2026 EO, DeepSeek-related device bans) stops short of removing general public access. NOT FOUND: No evidence of any new formal US government action between 2026-07-03 and 2026-10-08 that removed general public access to any major AI model (OpenAI, Anthropic, Google, Meta, xAI, Microsoft, Amazon, Mistral, DeepSeek, Alibaba, ByteDance, Moonshot, Zhipu); no passed federal legislation banning DeepSeek or any other model for the general public; no pending bill with a scheduled vote before Dec 31, 2026 found.

- [2026-06-12] An actual precedent occurred this year: the US Commerce Department issued an export control directive ordering Anthropic to suspend access to Claude Fable 5 and Claude Mythos 5 for all foreign nationals; on June 12, the U.S. government sent Anthropic a letter prohibiting access to both Mythos 5 and Fable 5 to any non-U.S. national whatsoever, regardless of their location, due to national security concerns, and as a result, Anthropic revoked access to both models for all customers. This shows the reference class event (formal US action → general public removal of a major US model) has already happened once in 2026, satisfying the literal definition for a brief period. https://en.wikipedia.org/wiki/Claude_Mythos - [2026-06-30] That precedent was reversed quickly: the U.S. Department of Commerce lifted its export controls on both models on June 30, 2026, Anthropic announced on X, ending a 19-day shutdown that froze access for enterprise customers across dozens of global cloud platforms. This establishes a base-rate "half-life" for such actions: roughly 2.5–3 weeks before reversal under industry/political pressure. https://www.marketscale.com/industries/software-and-technology/us-lifts-export-controls-on-anthropics-claude-fable-5-and-mythos-5-ending-19-day-shutdown - [2025-02-07 to ongoing] The main other reference-class track is the DeepSeek saga: H.R.1121, the "No DeepSeek on Government Devices Act," was introduced in the House February 7, 2025, "to prohibit the use of DeepSeek by the executive agencies, and for other purposes." As of the most recent check, the bill's status remains "Introduced," having only been referred to committee, with no further action recorded. Over roughly 20 months this bill has not advanced to passage, illustrating how slowly comparable "ban" legislation historically moves. https://www.congress.gov/bill/119th-congress/house-bill/1121 - [2026-01-14 (effective date in FY2026 NDAA, P.L.119-60)] The only enacted federal action concerning a Chinese AI model targets government/contractor systems, not general public access: Section 1532 prohibits DOD from using or acquiring covered AI systems—including those from DeepSeek and High Flyer—or systems from covered nations, with the SECDEF able to grant case-by-case waivers for research, training, evaluation, or national-security-related military activities. A companion intelligence-community provision similarly only restricts official systems: Section 6604 instructs the Director of National Intelligence to create guidelines that require the removal of DeepSeek, or its successors, from IC and IC-related systems. Neither action reaches ordinary consumer/public channels, so neither would qualify as "general public removal." - [2026-05-26 snapshot] Independent trackers confirm no general public ban exists: as of May 26, 2026, no nationwide U.S. consumer ban was identified in the official sources reviewed; DeepSeek is still generally accessible for ordinary personal use, but it is restricted in important government, defense, intelligence, state-government, contractor, and institutional/workplace contexts. https://chat-deep.ai/privacy-security/is-deepseek-banned-in-us/ - [2026-07-07] Signaling continued escalation risk but still only deliberation, not action: on July 7, 2026, Reuters reported exclusively that Chinese authorities held meetings with Alibaba, ByteDance, and Z.ai about potentially restricting overseas access to China's most advanced AI models, including models not yet released, and including open-weight releases — a reactive/mirroring dynamic to US actions, not evidence of a new completed US action. - [2026-07-22] A separate rumor track (distillation sanctions) remained unconfirmed: no Chinese AI models have been sanctioned as of July 22, 2026; on July 21, US Treasury Secretary Scott Bessent said the administration could sanction Chinese AI developers if it finds they were built by distilling American models, and promised action within days or weeks, with four labs named in the underlying allegations: Moonshot AI, DeepSeek, MiniMax, and Alibaba. No confirmed follow-through was found as of the search date. - [2026-07-03] A near-identical sibling Polymarket contract specifically for Chinese models opened the same day as this market: this market resolves "Yes" if the US federal government passes legislation, issues an executive order, issues an export control, or takes any other formal action that directly has the effect of generally removing public access within the US to a major Chinese AI model, between market creation and December 31, 2026; to qualify, ordinary public access to the model must be eliminated across ordinary public channels within the US, and removals not caused by formal government action will not qualify. A press summary of this sibling market put trader-implied odds at roughly a quarter chance as of late July 2026 (treat as a stale historical data point, not current pricing). - [2026-01-14] Outside-forecaster base rate from a retail prediction market specific to DeepSeek: a Manifold Markets question "Will DeepSeek be banned in US before 2027?" was trading around 5% as of early October 2026 per the related-questions panel, indicating low forecaster-assigned probability for a full consumer-facing federal ban materializing within the covered window (manifold is lower-volume/lower-accuracy than Polymarket but useful as an independent outside view). - [2026-07-06] Congressional Research Service analysis of the broader AI/defense legislative environment found no enacted measure reaching general public access: the FY2026 NDAA's AI provisions are confined to DOD use/acquisition restrictions with waiver authority, not public-facing bans , reinforcing that Congress's actual enacted output to date has consistently been narrower than "general public removal." BOTTOM LINE: The reference class shows this exact type of event has already happened once in 2026 (Fable 5/Mythos 5, June 12–30) via an export-control directive, but was reversed within 19 days, and all other federal actions to date (NDAA sections, the stalled "No DeepSeek" bill, state-level device bans) have stopped short of eliminating general public access to any major model; independent forecasters (Manifold ~5% for a DeepSeek ban, press-reported ~26% on the Chinese-model sibling market) imply moderate-to-low odds, suggesting a base rate meaningfully above zero but still well under 50% for a new, qualifying, sustained removal by the market's deadline. NOT FOUND: No evidence of any additional (second) full "general public access removal" episode for a major US-company or Chinese-company model since the June 2026 Fable/Mythos episode; no confirmed Treasury/Commerce sanction action against Chinese labs for distillation as of early October 2026; no official Polymarket resolution data for the sibling Chinese-model contract beyond the stale July 2026 press snapshot.

YES means: between 2026-07-03 and 2026-12-31, a US federal legislative, executive, or export-control action generally removes ordinary public access within the US to at least one flagship model from a named company (OpenAI, Anthropic, Google, Meta, xAI, Microsoft, Amazon, Mistral, DeepSeek, Alibaba, ByteDance, Moonshot AI, or Zhipu/Z.ai), even briefly. - [2026-06-12] The exact causal mechanism needed for YES already fired once: BIS used a non-public "is-informed" directive against Anthropic's Fable 5/Mythos 5 that caused a full US-and-global shutdown because the company "had no reliable way to verify nationality in real time," so it "suspended both models for all users." This shows the tool is live, tested, and easy to trigger again against any covered company. (https://www.theregreview.org/2026/10/06/oktenli-when-a-chatbot-becomes-an-export/) - [2026-10-03] The dispute that could trigger a second, Moonshot-targeted action has sharply escalated: Bessent told Axios Chinese firms are running "industrial distillation" and specifically named Moonshot's Kimi, claiming it "once sent People's Liberation Army weapons plans back to Anthropic" — a national-security framing (not just IP) that raises the odds of an Entity List or sanctions action rather than mere rhetoric. (https://huggingnews.com/ai/update-us-treasury-secretary-labels-chinese-ai-model-distillation-as-the-887284b2) - [2026-09-13] A follow-on technical audit found the underlying exploit "persisted for models hosted on Microsoft Azure," meaning any future action targeting the extraction/distillation pathway could sweep in a major US-company-hosted model (OpenAI/Anthropic models on Azure), not just a Chinese model, widening the range of models a single action could hit. (https://huggingnews.com/ai/update-us-treasury-secretary-labels-chinese-ai-model-distillation-as-the-887284b2) - [2026-07-22/23] The sanctions threat against Moonshot is specific and stated as imminent, not vague: Bessent warned that when PRC firms "conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table," with Moonshot's Kimi explicitly named as the target under review. (https://techcrunch.com/2026/07/22/treasury-threatens-sanctions-after-white-house-claims-moonshot-distilled-anthropics-fable/) - [2026-09-17] The administration's enforcement capacity for such actions is strengthening procedurally: the Senate Banking Committee voted 13-11 to approve Abby Warren as BIS assistant secretary for export administration, filling the key post that would execute any new AI-model export-control directive. (https://exportcompliancedaily.com/topic/ai_export_controls) - [2026-10-05] Congressional hawks are actively pressuring the administration to tighten, not loosen, China-facing tech controls: House Foreign Affairs Committee Republicans "criticized a reported deal by U.S.-based Oracle to lease 100,000 chips to Chinese technology firm Tencent," showing continued political appetite for aggressive China-AI actions that could spill into model-level restrictions. (https://exportcompliancedaily.com/article/2026/10/05/lawmakers-oppose-us-chip-deal-with-chinas-tencent-2610020004?BC=bc_6ac65a503d905) - [2026-10-02] A former State Department official's analysis argues the US retains real legal tools here, concluding "the U.S. may be able to take some steps to impose license requirements on certain dangerous uses" of open-weight models — i.e., credible legal pathways exist for Commerce to act again, distinct from pure rhetoric. (https://exportcompliancedaily.com/article/2026/10/02/former-us-official-analyzes-possibilities-obstacles-to-controlling-open-ai-models-2610010034?BC=bc_6ac65a503d905) - [2026-07-20] Axios reporting (cited in secondary coverage) found the administration had already revived shelved 2025 proposals including "Entity List additions, security advisories, and federal procurement rules" aimed at Chinese AI models — proposals that, if an Entity List designation is finalized against a company like Moonshot or Zhipu, would likely force US cloud/app-store delistings analogous to the June Anthropic shutdown. (https://vibraniumlabs.ai/blog/chinese-ai-models-sanctioned-whats-actually-happening) - [2026-07-24] Industry's own defensive mobilization — "an open letter led by Nvidia CEO Jensen Huang" ultimately drawing 132 signatories including Amazon, Google, Meta, Microsoft and OpenAI warning against "sweeping restrictions" — is itself evidence insiders view near-term broad US action against a major model as a real, non-trivial possibility worth lobbying against. (https://www.caixinglobal.com/2026-07-29/moonshot-open-sources-kimi-k3-as-us-china-ai-tensions-intensify-102468927.html) BOTTOM LINE: The precise mechanism required for YES (a Commerce/Treasury national-security or export-control directive forcing a company to pull a major model from general US access) has already worked once in 2026, and the same administration is now actively escalating a parallel, more serious national-security-framed case against Moonshot's Kimi (PLA-weapons-plan allegations as of Oct 3) with strengthened BIS enforcement leadership and sustained congressional hawkish pressure — all pointing to a credible, non-speculative path to a second qualifying event before year-end. NOT FOUND: No confirmed Entity List designation, OFAC sanction, or new BIS directive against Moonshot AI, Kimi, or any other named company as of 2026-10-08; no confirmed date for when Treasury's "coming days or weeks" review (promised July 21) will conclude.

YES means: the market's title event is affirmed — i.e., a formal US government action (legislation, executive order, export control, or similar) causes general removal of public access within the US to a major AI model (from the listed companies) at some point before Dec 31, 2026, 11:59 PM ET, with that removal still in effect on the resolution date. For this NO-case dimension, I argue why the market is likely to resolve NO. - [2026-10-03] As of the most recent dated reporting, Treasury Secretary Bessent's distillation-theft rhetoric against Moonshot/China remains at the accusation/threat stage: the Treasury sanctions already imposed signal the administration is willing to move beyond rhetoric is contradicted by the same-week primary account, which states only that Bessent "told Axios" firms are engaging in "industrial distillation" and that this follows "a period where Bessent observed China becoming more aware" — no sanction, Entity List action, or executive order is described as enacted (huggingnews.com, Oct 3, 2026). - [2026-07-22/07-27] The only concrete escalation path (Moonshot/Kimi K3 distillation allegations) has sat at "sanctions on the table" since July 21–27, 2026 with no confirmed follow-through nearly three months later; a dedicated tracker explicitly states no Chinese AI models have been sanctioned as of July 22, 2026... Bessent... promised action within days or weeks , and no subsequent source in this search confirms that promise was kept. - [2026-09-08] Even the most recent formal US agency product on distillation is an advisory, not an access-removing action: the CNN report describes a joint agency advisory naming a gray market of proxies known as "transfer stations"... singling out DeepSeek and Alibaba Group for knowledge-distillation campaigns — language, not a legal removal mechanism. - [2026-10-01/10-02] Congressional and executive-branch posture on China-AI controls remains deliberative: Moolenaar's Sept. 18 push for tighter "advanced AI products" export controls and a former State Department official's Oct. 2 essay arguing the US "may be able to" impose license requirements on open-weight models are both forward-looking proposals, not enacted law (exportcompliancedaily.com, Oct 1–2, 2026). - [2026-10-01] Chip-focused export-control legislation—the closest things to AI-restriction bills with momentum—remains stalled: Chip-Related Export Control Bills Remain in Senate Limbo , indicating Congress has not moved any AI-model-access bill to passage either. - [2025-06/2026-10] The two leading "ban a model for the public" bills have made zero procedural progress in over a year: S.2177 (No Adversarial AI Act) and H.R.4142 remain at "Introduced... Referred to Committee" status with no committee markup or floor action recorded as of the latest govinfo bulk-data status update (congress.gov, govinfo.gov). - [2026-07-27] Industry legal guidance written after the Fable/Mythos episode treated a new blanket model-access ban as speculative, not current: as of mid-2026, there is no blanket ban on sharing model weights internationally... Any of them could become policy in the second half of 2026 , framing all restrictive options as "watch for," i.e., not yet happened. - [2026-10-06] Legal scholarship published just two days before this report examined the June Anthropic export-control directive specifically because it was anomalous and controversial, concluding the reversal "left the underlying legal questions unresolved" — signaling BIS itself may be cautious about repeating a tool that drew criticism as "the first of its kind" and legally untested, raising the bar for a repeat before year-end. - [2026-05/2026-10] The only enacted federal statutory actions touching a "major AI model" (FY2026 NDAA Section 1532, IC guidance under Section 6604) are confined to government/contractor systems and explicitly allow waivers, reinforcing a consistent pattern of Congress stopping short of general-public removal even when it does act. - [2026-10-08] No search turned up any newly enacted executive order, statute, or Commerce/Treasury action between July 3 and October 8, 2026 that caused a major-model provider to generally disable a flagship model for the US public — the single historical precedent (Fable/Mythos, June 12–30) remains the only instance all year, and it predates and is explicitly excluded by this market's "another" framing. BOTTOM LINE: Nearly three months into the question window, every potential escalation path (distillation sanctions, Entity List action, chip/AI export bills, adversarial-AI acquisition bans) remains at the rhetorical, proposed, or stalled-in-committee stage; no second qualifying public-access removal has occurred since the market opened, and the one historical precedent was resolved in 19 days and falls outside the window, supporting a NO resolution absent a sudden new action in the final months of 2026. NOT FOUND: No confirmed Treasury/Commerce sanction, Entity List designation, or executive order enacted (as opposed to threatened) against any Chinese AI lab as of October 8, 2026; no scheduled floor vote for S.2177/H.R.4142 or any chip/AI export-control bill; no evidence of any second BIS directive analogous to the June 2026 Anthropic letter since market creation.