Year a Major Government Restricts an AI Release
Year a Major Government Restricts an AI Release
Experts give it until 2030 before the U.S., U.K. or EU first blocks an AI release on safety grounds.
Hover to highlight a group, click to show or hide itTap a group to show or hide it
Our Analysis
How do we interpret these results
Experts and superforecasters give closely similar forecasts for when a qualifying government intervention will first occur. The median expert and superforecaster each assign a 50% chance of resolution by 2030, while the public is a little later at 2032. The median expert gives a 25% chance of resolution by 2028 and a 75% chance by 2034, indicating a meaningful probability that such an intervention occurs within the next decade. Experts and superforecasters do not differ significantly across any elicited percentile. The public, by contrast, is later than experts: expert-public differences are statistically significant at the 5th, 25th, 50th, and 75th percentiles, while the groups are much closer at the 95th percentile.
On June 12, shortly after the survey closed, the U.S. federal government issued an export control directive suspending all access to Anthropic’s Mythos 5 and Fable 5 models by any foreign national. However, given these models had already been made publicly available, this directive does not meet our resolution criteria that such a direction must “delay, seriously restrict or condition the release of an AI system prior to its availability to the general public, citing safety or national security risks.”
Forecasts by group
The year by which a typical forecaster gives the event each chance of having happened
| group | 5% | 25% | 50% | 75% | 95% |
|---|---|---|---|---|---|
| Experts | 2027 | 2028 | 2030 | 2034 | 2040 |
| Superforecasters | 2027 | 2028 | 2030 | 2035 | 2045 |
| Public | 2027 | 2029 | 2032 | 2036 | 2043 |
Click any cell to compareTap any cell to compare
* Denotes non-overlapping bootstrapped confidence intervals between group aggregates.
Rationale analysis
The main lines of reasoning driving the spread in estimates
The Mythos precedent
Both early- and late-resolution respondents cite Anthropic’s voluntary restriction of Claude Mythos Preview but often draw opposite conclusions from the incident. Early-resolution respondents tend to read it as evidence the dangerous-capability threshold has arrived and that involuntary restrictions will soon follow: “Mythos just gave national governments a reason to take action this year.” Whereas many late-resolution respondents read the same episode as proof that voluntary restraint is an adequate substitute for state action, with one writing that “AI companies are likely to cooperate in order to avoid the precedent of binding restrictions,” and another that “there won't be any need as AI companies will decide not to deploy extremely powerful models just for mere economic reasons.”
The EU AI Act and its August 2026 enforcement powers
A frequent dividing line is whether the EU’s existing machinery is a ready vehicle for a qualifying action. Early-resolution respondents often treat it as such: “The EU AI Office gains explicit legal authority under Article 93 of the AI Act to restrict market availability of models posing systemic risks starting August 2, 2026—the earliest plausible trigger.” Some late-resolution respondents, however, argue the AI Act isn’t a good vehicle for imposing a pre-release restriction: “Even the AI Act doesn't actually give this power yet.”; “The question is whether [the AI Act] ever materializes as a halt, rather than a compliance demand.”
Inciting incident
Many respondents, representing both poles, agree a notable safety incident is a likely path to resolution, with several forecasting that an incident involving a released model could result in restrictions on future models of the same class. But differences emerge on whether such an incident is imminent: “Once one major AI incident or near-miss occurs, at least one of these three jurisdictions will act and I expect this to happen within roughly 2 years,” writes an early-resolution respondent, while a late-resolution respondent writes, “I don't expect this to happen until there is an AI-caused catastrophe,” and "without that trigger institutional inertia wins.”
The US political environment
Early-resolution respondents often treat the reported Trump executive order [it was signed on the last day the survey was open] as revealed intent: “Given that we see the Trump administration—which is highly anti-regulatory—considering this option, I think we will see more rational actors considering and implementing it relatively soon.” Another adds that it will be “good politics (voters are genuinely worried about AI.)” Late-resolution respondents tend to emphasize that the order was voluntary and [a stronger version] was ultimately shelved, and that meaningful U.S. action on this front will likely be deferred to a future administration: “Trump just backed away from AI regulation. So if the U.S. does it, I think it will happen after Trump leaves office”; “Unlikely in U.S. until after Trump.”
Capability trajectory
Many early-resolution respondents argue that frontier capabilities are escalating rapidly, increasing both the risk posed by these models and the likelihood of restrictions being imposed: “I expect a qualifying government intervention to be more likely than not…mainly because frontier model cyber/bio/autonomy risk are becoming legible to governments…”; “Pandora’s box is about to be opened. The latest models by several companies already are far too dangerous.” Late-resolution respondents tend to downplay the risk: “AI has not yet done anything that is truly alarming or existentially disastrous.”
Resolution-criteria threshold
Many early-resolution respondents read the bar as low: “The resolution criteria doesn't require the regulation to be particularly stringent, e.g. an order requiring the voluntary 2-3 month delay that Anthropic imposed on itself would count.” Others point to similar government restrictions: “Emergency action against a specific product can be near-instant once a galvanizing event occurs: the FAA grounded the 737 MAX three days after the second crash.” Whereas some late-resolution respondents read the same text as demanding a narrow, historically unprecedented form of action: “The resolution criteria are very narrow. General AI regulation, EU AI act compliance, transparency rules, or safety reporting would not count,” writes one, with another noting that “governments usually regulate deployment and use, rather than pre-release model publication.”
Geopolitical and economic pressures
Several late-resolution respondents argue that the race between the U.S. and China for AI supremacy, paired with influence the AI industry exerts over politics, will stymie action: “There's a lot of money (and thus lobbying) involved, plus there's a lot of international competition where regulation would probably backfire or be a leaky bucket problem.”; “Too many rich and powerful men expect to make a lot of money off of AI, and they tend to wield power over elected officials.” Several early-resolution respondents point to the potential for the national-security element to cut the other way: “The most likely trigger is not broad safety regulation, but a U.S. national-security-framed restriction on frontier model weights—something closer to an export-control measure designed to prevent adversaries from accessing, copying, or distilling cutting-edge models.”
Rationale examples
High-quality rationales from forecasters across the range
Early-resolution respondents
I do not expect this to emerge from calm, mature policymaking where everyone reads the papers, thinks deeply, and produces a beautifully balanced regulatory mechanism. That would require a different species. My central model is that the first binding intervention comes after an ‘oh shit’ event: a frontier model either materially accelerates cyber offense, discovers high-value zero-days at scale, enables ransom/extortion workflows, or appears plausibly useful for nation-state-level disruption. Mythos-like capabilities already seem to be pushing regulators toward the edge of the pool: Anthropic limited release voluntarily, the UK AISI found significant cyber capability improvements, and financial regulators are already treating this as a systemic cyber-risk issue rather than normal software launch drama.
China is already heading in the direction of controlling AI. What China considers unsafe may not exactly match what the US, UK and EU consider unsafe, and indeed the US, UK and EU may each differ in what they consider unsafe. The US tends to look for proof that something is unsafe before restricting its use, while the EU looks for proof that something is safe before allowing its use. So I would expect the EU (and UK) to act first (after China) on AI safety and the US to follow (maybe after an unsafe AI system has caused harm).
Considering just the US, there have been multiple times that technology was restricted for government use. The obvious one is nuclear technology—the US passed its first act regulating the nuclear industry in 1946. So—a year after using nuclear weapons. Satellite imaging was also not publicly available initially, and the publicly available resolution has lagged state of the art. There was a directive earlier this year restricting access to imaging data for regions of the Middle East.
Safety testing regulations are a standard requirement for many goods. For example, televisions and routers have to comply with electromagnetic compatibility rules and undergo testing against these requirements. Aircraft equipment and pharmaceuticals have more stringent testing requirements. Currently, AI is outside these rules and regulations and testing for new foundation models is conducted on a voluntary basis. The existence of these voluntary arrangements could be viewed as a first step toward directive testing and more formal regulation. Often formal rules and supporting legal frameworks are put in place only after there is clear evidence of harm. The 1938 Food, Drug and Cosmetic act was drafted and passed in response to a new product based on Sulfanilamide that killed over a 100 people.
Late-resolution respondents
I think it is highly unlikely we will see an event that will trigger [an] affirmative [resolution] on this question until there is a full AI evaluation regime in place. Otherwise, how would any given jurisdiction have proper evidence to delay or ban a model from public release? As a baseline, no country has such a review process at this time, nor is any country contemplating developing one. This is a major hurdle, because such a regime would take considerable effort to establish if it was to really evaluate an AI. It would require staff with significant technical expertise and the authority to carefully evaluate any AI before release. In practice, setting such a review process would take 1-2 years if we started today.
The question rewards a specific regulatory form: a pre-release binding action on a named system. Governments may instead regulate through procurement rules, export controls, or post-deployment liability. If they do, this question resolves later than the underlying question of whether AI is being meaningfully constrained by government.
I think court orders or injunctions are unlikely to delay, seriously restrict, or condition the public release of AI systems based on safety risks, because litigation is reactive. There needs to be someone who has a standing in the case of getting an AI system to be blocked, which seems pretty difficult to pull off ex ante - there's no argument of the form "this model will specifically harm a certain person in specific ways.
The U.S. government has tools (e.g. Secrecy Orders, National Security Letters, and classified procurement pipelines) to allow it to keep such suppression secret. So it is possible that some such suppression has already occurred, and/or that some will occur in secret in the future. It is, in fact, more effective for the U.S. government to suppress AI secretly, because public disclosure of the act of suppression will likely lead to increased efforts to steal the model. So this argument somewhat reduces the probability of a publicly acknowledged suppression.