The US government has restored the UK AI Safety Institute's access to the latest American AI models for safety testing, according to reporting by Tony Diver. The reversal comes roughly a month after Washington ordered OpenAI and Anthropic to stop sharing their models with the British regulator.
The earlier restriction had blocked UK officials from reviewing frontier systems built by Anthropic, OpenAI, and Google DeepMind. The UK AISI now has its access back.
What the freeze actually covered
When the US first moved to cut off model sharing, the order landed on OpenAI and Anthropic. Google DeepMind wasn't named in that initial directive, but its models were caught up in the wider block on UK regulatory review of frontier systems. All three of the major US AI developers were effectively off-limits to British safety testers.
📊 Market Data Snapshot
That's a meaningful gap. The UK AI Safety Institute was set up specifically to run pre-deployment evaluations on frontier models, and it can't do that work if the labs won't hand over access. For about a month, the institute was operating with a hole in its testing pipeline covering the most capable models on the planet.
A short-lived policy, a quick reversal
The turnaround came fast. A restriction imposed last month is now gone. That speed matters. It suggests the original order wasn't a carefully considered piece of long-term policy so much as a misstep that got corrected once the right people pushed back.
There's no public explanation yet for why the US changed course, or what conditions were attached to restoring access. The facts don't say whether new safeguards were put in place, whether the sharing arrangement is narrower than before, or whether the reversal is permanent. All we know is that the door is open again.
Why this isn't just an AI story
Crypto's AI narrative has been waiting for exactly this kind of signal. The sector's pitch — decentralized compute, data privacy, model training on open networks — depends on a regulatory environment that isn't actively hostile to cross-border AI collaboration. A fragmented regime where the US and UK can't even share models for safety testing is a bad backdrop for those projects.
So the restoration of access removes one specific uncertainty. It doesn't hand AI tokens a rally, and it doesn't change anything about how the SEC or CFTC treats crypto. But it does suggest the US isn't retreating into isolation on frontier AI. That's a small positive for the broader risk narrative, and it reinforces the slow grind toward regulatory clarity that institutional investors keep saying they need.
The two-tier problem nobody's talking about
Here's the thing about a restored access deal: it cuts one way. US models get tested in the UK. That doesn't mean UK models get reciprocal treatment in the US, and it doesn't mean the two countries are converging on a shared safety framework. What it institutionalizes is a lopsided arrangement where American labs benefit from British scrutiny while British developers still face an uneven path into the US market.
For crypto-AI projects, that creates a strange incentive. Tokens tied to projects that can demonstrate compliance with both US and UK safety standards may start to look like safer bets than single-jurisdiction plays. It's early, and the facts don't show any token actually doing this yet. But the logic is straightforward: if the regulatory landscape is bifurcated, the projects that can navigate both sides capture the larger market.
Watch for whether the restored access comes with any formal framework attached. Without one, this is a handshake, not a policy. And handshakes get reversed.
