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

1. Planner Fable 5.1 · Fermi decomposition + tool plan

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

Sub-questions (Fermi decomposition)

What specific US government action already removed public access to an Anthropic model in 2026 (the 'first' removal implied by 'another'), what legal mechanism was used, and is that mechanism still active and reusable?
What is the current state of the Anthropic–federal government relationship (e.g., Pentagon/DoD supply-chain-risk designation, disputes over usage policies, AI Action Plan compliance) as of mid-2026, and are officials signaling further action?
Are there pending bills, executive orders, or Commerce/BIS export-control rules (e.g., frontier model diffusion rules, restrictions on model provision to foreign persons) that could plausibly force Anthropic to pull a model from general US availability before Dec 31, 2026?
How has Anthropic responded to prior actions—did it fully remove a model across all public channels (API, Claude.ai, cloud platforms), or only partially—and would it likely do so again given the resolution criterion requires removal across ordinary public channels?
What is the base rate of the federal government taking formal actions removing public access to specific commercial AI models (e.g., DeepSeek bans, TikTok-style actions), and how often has it happened to a domestic company?
What do the Polymarket price history and related Polymarket/Kalshi markets (Anthropic bans, AI regulation, export controls) imply about crowd expectations, and is there disagreement between venues?

Planner reasoning

The title implies a prior US government action already removed public access to an Anthropic model in 2026, so the key questions are what that precedent was, whether the underlying legal mechanism (e.g., supply-chain-risk designation, export control, EO) is ongoing and likely to be applied again, and how Anthropic's relationship with the administration is trending. The Polymarket price is the primary anchor; news search on the Anthropic-government dispute and any pending legislation or executive actions will refine it.

2. Research Parallel tool calls

tool	status	time	items	summary
polymarket_direct	OK	0.3s	1	## This Market's Polymarket Data US Government removes public access to another Anthropic AI model in 2026? - Current price (probability): 10.50% - 7-day price change: +3.50% - 30-day price change: -10.00% - Total volume: $34,019 (USD notional) - Price range: 6.50% - 33.50% - Data points: 70 da
polymarket_related	OK	3.2s	0	Scanned 100 active Polymarket markets, kept 0 matches. keyword 'Anthropic': 0 markets \| keyword 'Claude': 0 markets \| keyword 'AI model ban': 0 markets \| keyword 'export control AI': 0 markets
kalshi_related	OK	3.0s	2	2 related markets / summaries. keyword 'Anthropic': ok \| keyword 'Claude': no matches \| keyword 'AI ban': ok
claude_news	OK	18.2s	15	## Key Findings The 2026 removal event (already occurred, referenced by the question's "another"): - On June 12, 2026, the U.S. Commerce Department ordered Anthropic to immediately prevent all foreign nationals from using its two most advanced large language models, disabling Claude Fable 5 an
article_search	OK	1.0s	28	Fetched 28 unique articles across 5 queries (lookback=120d). 'Anthropic government removes access Claude model': 10 articles \| 'Anthropic Pentagon supply chain risk designation': 10 articles \| 'Anthropic executive order ban Claude': 10 articles \| 'export controls frontier AI models Commerce BIS 2026
gdelt_news	OK	120.4s	20	GDELT: 20 articles across 3 queries (lookback=90d). 'Anthropic Claude ban government': 10 hits \| 'Anthropic supply chain risk': 10 hits \| 'Anthropic export control': error GDELT rate-limited after retries (429)
congress_bills	OK	0.6s	0	Found 0 bills matching 'Anthropic artificial intelligence model restriction'.
congress_bills	OK	1.2s	7	Found 7 bills matching 'frontier AI model export control'.
wikipedia	OK	0.1s	2	Fetched 2 Wikipedia entries (0 missing pages).

3. Evidence Brief Sonnet 5 · 7892 chars

# Current state The "first" removal already happened and was reversed: Commerce/BIS export-controlled Claude Fable 5 and Mythos 5 in June 2026, cutting off all US public access for ~2.5 weeks before restoring it by June 30, 2026. Separately, the Pentagon's "supply-chain risk" designation (Feb 2026) was struck down by a federal judge (Aug 28, 2026) as unlawful retaliation — that dispute never removed *public* access, only government/contractor use. As of ~Sept 2026, no Anthropic model is currently blocked from general US public access; the question asks whether a *second, new* removal action occurs before Dec 31, 2026. # Timeline of key events - 2026-02-27 (confirmed): Trump orders federal agencies to cease all use of Anthropic tech; DoD given 6-month phase-out. [Wikipedia, claude_news] - 2026-02 (confirmed): Pentagon (Hegseth) labels Anthropic a "supply chain risk" after dispute over unrestricted military use (autonomous weapons, mass surveillance). [Wikipedia] - 2026-03 (confirmed): Federal judge issues preliminary injunction blocking the supply-chain-risk designation. [Wikipedia] - 2026-06-02 (confirmed): Executive order directs agencies to build a process for benchmarking/assessing new AI models before "wide release." [claude_news] - 2026-06-12 (confirmed): Commerce Dept. issues export-controls directive; Anthropic disables Claude Fable 5 and Mythos 5 for all US users within hours, citing national-security/jailbreak concerns (cyberattack risk). [TechCrunch, CBS, CNBC] - 2026-06-26 (confirmed): Commerce (Lutnick) permits Mythos 5 release to "trusted partners" only (100+ vetted US companies/agencies) — not general public. [CNN, TechCrunch] - 2026-06-30 (confirmed): Commerce fully lifts restrictions after Anthropic strengthens safeguards (with benchmark score degradation); Fable 5 restored globally on Claude.ai, API, AWS/GCP/Azure by July 1. [CBS, CNBC, BBC] - 2026-08-28 (confirmed): Federal judge (Rita Lin) permanently voids the Pentagon supply-chain-risk designation as unlawful First/Fifth Amendment retaliation; narrower D.C. Circuit case still pending. [TechCrunch, Fortune, Wired] - 2026-09 (confirmed, unrelated to bans): Anthropic discloses a 4th accidental internet-access incident during red-team testing — not a government action. [CBS Colorado] # Event Will the US federal government take a NEW formal action in 2026 (after the June Fable/Mythos episode) that generally removes US public access to an Anthropic AI model? # Outcomes to forecast - Yes — another formal US government action removes general public access to an Anthropic model before 2026-12-31 - No — no such second removal occurs # Kalshi market anchor No direct Kalshi YES price was returned in raw research for this ticker (kalshi_direct tool output absent). Best available cross-venue proxy: **Polymarket price = 10.5% YES** (as of data pull), up +3.5% over 7 days but down -10pp over 30 days; range 6.5%–33.5% over 70 days; volume modest ($34K total). Treat this as the working consensus anchor pending confirmation of the actual Kalshi price. # Sub-question answers 1. **First removal mechanism**: Commerce Dept./BIS issued an "export controls directive" (national-security authority) on 2026-06-12 forcing Anthropic to disable Fable 5/Mythos 5 for all US users within hours; lifted 2026-06-30. This exact mechanism (BIS export control citing national security) is legally established and reusable against any future frontier model. [TechCrunch, CSIS, JustSecurity] 2. **Anthropic-federal relationship**: Tense but partially resolved — Pentagon's supply-chain-risk label voided by court (2026-08-28), a narrower case still pending in D.C. Circuit; Commerce/BIS retains export-control leverage; a June 2026 EO calls for pre-release benchmarking of new frontier models, signaling continued scrutiny of future releases. [Wikipedia, techcrunch, CRS] 3. **Pending legislative/regulatory triggers**: No bill specifically targeting Anthropic found (0 hits). General frontier-AI/export-control bills exist (China AI Power Report Act, NDAA FY2027) but none confirmed to mandate model access removal. [congress_bills] 4. **Anthropic's prior response**: Full removal occurred across ordinary public channels (Claude.ai, API, cloud platforms) within hours when ordered — satisfying this question's resolution bar; company complied fully rather than resisting. [TechCrunch, CNBC] 5. **Base rate**: This is a novel, unprecedented action against a domestic US company (prior model bans like DeepSeek targeted foreign firms); one such domestic incident already occurred in 2026, suggesting elevated (not negligible) recurrence risk given active EO/benchmarking regime and unresolved BIS authority. [Wikipedia, CSIS] 6. **Cross-market signals**: Polymarket prices this at 10.5%, well below its 30-day high (33.5%), suggesting fading probability after the June episode resolved without recurrence; no dedicated Kalshi-related Anthropic-ban markets found beyond IPO-race and US-stake markets (93%/19%), which are not directly informative here. # Key facts (high-confidence, factual) 1. [TechCrunch/CBS/CNBC] Commerce/BIS forced full US shutdown of Fable 5/Mythos 5 on 2026-06-12; restored by 2026-06-30. 2. [Wikipedia/Fortune] Pentagon's Feb 2026 "supply-chain risk" designation was judicially voided 2026-08-28; a narrower case remains pending. 3. [Polymarket] Current probability 10.5%, down from a 33.5% peak, up slightly in the last week. 4. [congress_bills] No bill specifically targeting Anthropic model access found; general AI/export bills pending but not model-specific. # Cross-market signals - Kalshi related: OpenAI/Anthropic IPO-first market (93% Anthropic) and US-stake-in-both market (19%) — not directly predictive of a ban. - Polymarket: 10.5% YES, declining from 33.5% peak, low volume ($34K) — thin market, moderate conviction against recurrence. - Sportsbook implied: N/A. # Analyst opinions and speculation - CSIS/JustSecurity: legal basis for the June export-controls directive is contested/novel, but precedent now exists for BIS to act again quickly against frontier models. [CSIS, JustSecurity] - The Conversation: US government now effectively "controls global access" to frontier AI via export-control leverage, implying recurring risk as new, more capable models launch. [The Conversation] - Claude_news synthesis: "balanced trader sentiment" reflecting genuine uncertainty given ongoing friction and the June 2026 EO mandating pre-release capability benchmarking. # Directional lean per outcome - **Yes**: Supported by precedent (already happened once via BIS in 2026), unresolved EO benchmarking regime, ongoing distrust between Anthropic and administration (esp. on frontier/cyber-capable models), pending D.C. Circuit case. - **No**: Supported by court rulings favoring Anthropic (Aug 2026), swift resolution/de-escalation of the June episode, Anthropic's proactive safeguard strengthening reducing trigger risk, declining Polymarket odds, thin evidence of new imminent legislative/BIS action specifically targeting Anthropic before year-end. # Gaps / unknowns - No live Kalshi YES price captured for this specific ticker in raw research — must be verified separately. - Unclear whether future Claude releases (e.g., Opus successors) will trigger similar national-security review under the June EO's benchmarking process. - Outcome of pending D.C. Circuit appeal on Pentagon rule unknown; could reintroduce risk if reversed. - No visibility into internal Commerce/BIS deliberations post-June for new triggers (e.g., new jailbreaks, cyber incidents). # Calibration anchors - Polymarket YES price: 10.5% (anchor proxy, thin volume, declining trend). - Precedent: one full-model removal-and-restoration cycle already occurred in 2026 (~18 days duration), showing the mechanism is fast-acting and repeatable but also historically short-lived once resolved.

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

claude-fable-5-1

52% conf

Yes	12%
No	88%

The only available consensus anchor is Polymarket at 10.5% YES (no direct Kalshi price captured), declining from a 33.5% peak after the June Fable 5/Mythos 5 BIS episode resolved in ~18 days. The mechanism is proven and reusable, and the June EO's pre-release benchmarking regime plus ongoing administration-Anthropic friction keep tail risk alive if a new frontier model launches before year-end. However, Anthropic strengthened safeguards and complied quickly, courts have twice sided with Anthropic (Aug 28 voiding the Pentagon designation), no legislation targets Anthropic model access, and only ~3.5 months remain. I land slightly above the market at 12% to reflect the demonstrated precedent and unknown new-model release timing.

gpt-5.6-sol

64% conf

Yes	11%
No	89%

The 10.5% Polymarket proxy is the best available consensus anchor because no direct Kalshi price was captured. The June BIS action establishes a reusable mechanism, and Anthropic's rapid compliance means another directive could quickly produce a qualifying removal. However, the prior restrictions were lifted after safeguards improved, the Pentagon designation was judicially voided, and no model-specific bill or imminent BIS action has surfaced. These factors support staying close to the anchor, with only a small upward adjustment for the active benchmarking regime and continuing federal-Anthropic friction.

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

Both forecasts converge tightly around the Polymarket anchor (12% and 11%) despite explicitly noting it's a thin, low-volume proxy ($34K total volume) for a *different* venue than the actual Kalshi contract being forecast — neither attempts an independent base-rate estimate that might diverge meaningfully from this proxy, effectively anchoring twice on the same weak signal.

Neither forecast seriously grapples with sub-question 5's base-rate point that "one such domestic incident already occurred in 2026" within just the first ~9 months, which — if treated as a Poisson-style recurrence rate rather than a one-off — could imply a higher full-year probability of a second event than ~11-12%, especially with ~3.5 months of exposure remaining under an active, reusable BIS mechanism.

Both forecasts underweight the "new model launch" trigger pathway: the brief flags that Opus successors or other new Claude releases could trigger the June EO's pre-release benchmarking review, and neither forecast discusses Anthropic's release calendar for Q4 2026, which is the single most concrete forward-looking catalyst mentioned in the gaps section.

Neither forecast addresses the pending D.C. Circuit appeal on the Pentagon designation as a distinct tail risk; if reversed before year-end, this could reinstate an action, but critically the brief notes the Pentagon action only ever restricted government/contractor use, not general public access — both forecasters correctly seem to implicitly exclude this, but neither explicitly clarifies why this pathway doesn't count toward "Yes," leaving ambiguity in their reasoning.

Both forecasts treat the 30-day decline from 33.5% to 10.5% as pure de-escalation signal, but neither considers that thin volume ($34K) makes this price move potentially noise-driven rather than informative, meaning the "declining trend" evidence may be overweighted relative to its statistical reliability.

Forecaster 2 claims higher confidence (0.64) than Forecaster 1 (0.52) despite producing an almost identical probability and citing nearly identical evidence — this confidence divergence is unexplained and suggests inconsistent calibration methodology between the two rather than genuinely different information processing.

6. Aggregator Fable 5.1 · final calibrated forecast

14% (58% confidence)

Yes: 14% No: 86%

Both forecasters cluster tightly at 11-12% around a thin Polymarket proxy, and the critique is right that this double-anchors on a weak signal and underweights the recurrence argument: a reusable BIS mechanism already fired once in 2026, Anthropic complies within hours, and a Q4 frontier release under the June EO benchmarking regime is a concrete catalyst that could trigger a second review. However, a naive Poisson extrapolation overstates risk because the June episode was tied to specific jailbreak concerns that Anthropic addressed with strengthened safeguards, the government de-escalated within 18 days, courts have repeatedly sided with Anthropic, and the pending D.C. Circuit case concerns government/contractor use rather than public access and so cannot resolve this Yes. With only ~3.5 months of exposure remaining and no identified imminent BIS or legislative action, I nudge modestly above both forecasts and the market to 14% to reflect the demonstrated precedent and new-model trigger pathway, while keeping No as the strong favorite.

Pipeline Timing

Total pipeline time: 208.9s

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