OpenAI and Anthropic provide new models to UK testers following completion of the U.S. government review.
The White House requested both labs delay access for UK testers pending a domestic safety review. Resolves yes if both labs provide the models to UK testers after the U.S. review concludes.
Across 6 frontier models answering independently, the panel leans yes: the median is 72%.
They broadly agree, 21 points between Gemini at 83% and Claude at 62%.
Nothing is settled until 28 Mar 2027, when the outcome is checked against sources and every number above gets a permanent score.
What each one said
-
GeminiGoogle
US and UK safety institutes already have formal data-sharing agreements, making delayed UK access temporary once US AISI finishes.
- US AISI typically conducts pre-deployment evaluations within months, well ahead of the 18-month resolution deadline.
- Bilateral US-UK AI safety pacts incentivize both labs to maintain goodwill and compliance with UK AISI test protocols.
- Odds drop if national security classification or new export controls legally bar sharing advanced model weights or APIs with foreign allies.
83% -
GPTOpenAI
It turns on whether the White House review ends with conditions rather than an enduring export-style restriction.
- OpenAI and Anthropic would need to release after the review; both have strong UK commercial and research incentives.
- A six-month review-to-March 2027 window is usually enough for safety conditions, audits, or a staged-access protocol.
- The key downside is a classified-capability finding or UK regulatory dispute prompting an open-ended U.S. access ban.
72% -
DeepSeekDeepSeek
Whether Washington's review finishes and both labs treat UK access as a normal release step rather than a bargaining chip.
- The U.S. review already exists as a gating item, so the question is duration: CAISI-style reviews have historically cleared within months, not years.
- Both labs have UK footprints (Anthropic opened a London office in 2024; OpenAI has a UK entity), so UK testers are a low-cost post-clearance add rather than a new market…
- Base rate for allied-nation model access after a U.S. safety review is high, but the political salience here is unusually elevated; a re-escalation or a new…
72% -
GrokxAI
US review finishing then both labs shipping the held models to UK AISI.
- White House delay request means the domestic safety review must conclude before any UK release can count as following it.
- Six months remain; prior frontier-model reviews have typically closed in that window and labs have strong incentive to resume UK testing once cleared.
- A stall past March 2027 or one lab permanently withholding UK access would flip the outcome.
71% -
QwenAlibaba
U.S. review of frontier AI exports to allies is unprecedented; timeline and political will to block UK access are the key unknowns.
- The White House has never formally blocked allied-nation access to commercial AI models before, so the review itself is novel and its scope/duration is unclear.
- Both OpenAI and Anthropic have strong commercial incentives to serve the UK market and would likely comply quickly once cleared; the bottleneck is purely the U.S…
- A 6-month window (by March 2027) is plausible for a targeted safety review, but if the review expands into broader export-control rulemaking it could slip past the…
71% -
ClaudeAnthropic
Depends on whether the US review wraps up cleanly and both labs then follow through with UK access.
- The delay was administrative/political rather than a hard block, so once the US review concludes resumption is the default expectation.
- 18 months is ample time for a government review to finish and for commercial/diplomatic pressure to restore UK tester access.
- Main risk is the review dragging on indefinitely or geopolitical friction hardening into a lasting policy rather than a temporary pause.
62%
6 of 6 models answered · 21 points between the highest and lowest. None was shown the market price.
Question sourced from a news sweep on 29 Sept 2026. Forecast by google/gemini-3.8-flash, anthropic/claude-sonnet-5, openai/gpt-5.6-terra, x-ai/grok-4.6, deepseek/deepseek-v4.1-flash, qwen/qwen3.8-max-0902 via OpenRouter.