Resolution criteria:

This market resolves YES if, before January 1, 2027, OpenAI or xAI is legally required by a government body to restrict access to any general-purpose large language model (chat/completion model intended for broad use across many tasks, not a narrow fine-tune for a single domain, e.g. GPT-5.5-Cyber, GPT-Rosalind, etc.) on security or safety grounds, where the restriction reduces the population of users who can access the model relative to its prior or planned availability. This applies to both already-released models and models that were undergoing testing/pre-release at the time of the directive. National security is one possible basis for such a mandate, but the criteria are not limited to it — restrictions justified on grounds such as public safety, AI safety risk, cybersecurity risk, or similar government-stated rationales also qualify. This includes, but is not limited to, restrictions based on nationality/citizenship.

Qualifying events include:

• A government directive (US or otherwise) requiring the company to disable or restrict access for non-citizens/non-nationals of a specific country, similar in kind to the Fable 5 / Mythos 5 suspension.

• A government directive requiring the company to suspend or restrict access to the model for all users (i.e., a full or near-full suspension pending review), justified on security or safety grounds.

• A government directive resulting in the model being restricted to a vetted/licensed subset of users (e.g., approved government agencies, approved enterprise partners, or a formal allowlist), where the directive is grounded in a security or safety rationale (national security, cybersecurity, AI safety risk, public safety, etc.) and is a legal requirement, not a recommendation the company chose to follow.

• A legally mandated restriction on a model that has not yet been publicly released — e.g., if OpenAI is legally required to withhold or restrict release of a model still in testing, based on a security/safety finding by a government body (such as CAISI or UK AISI).

• Any access tier explicitly gated by nationality/citizenship verification (as opposed to geographic IP-based blocking (such as the already existing restrictions for certain countries), which does not by itself qualify — see exclusions).

Does NOT qualify (resolves NO on these alone):

• A company voluntarily delaying or restricting release of a model in response to a finding, recommendation, or informal request from a government body (e.g., CAISI, UK AISI, or any agency) where compliance was not legally required. Example: if OpenAI states "we didn't release this model because CAISI found X," and there was no legal obligation to act on that finding, this does NOT count — even if the company explicitly cites the government finding as its reason.

• Standard geographic availability restrictions (e.g., a model not being launched in certain countries/regions at all, GDPR-driven EU feature limitations, or IP-based geoblocking without nationality verification).

• Restrictions targeting only nationals of countries already subject to broad US export controls as of [date market created] (e.g., existing China/Russia restrictions under prior policy) — the bar is a new restriction extending beyond the pre-existing sanctioned-country list.

• A company's own purely-internal safety decision to limit a model's release, where there is no government directive involved at all.

• Restrictions imposed for purely non-security/non-safety reasons (e.g., a directive based solely on antitrust, trade/tariff disputes unrelated to security, or intellectual property concerns).

• A model that is, by its own naming, marketing, or system card, scoped to a narrow specialized purpose (e.g., a hypothetical "GPT-5.5-cyber" variant explicitly marketed as a cybersecurity-specific tool, not the general chat/API model). Standard API variants (mini, pro, turbo, etc.) do not count as specialized variants.

• Account-level bans, fraud/abuse enforcement, or sanctions-list compliance (e.g., blocking a specific sanctioned individual or entity) that aren't part of a broader access policy change.

• Restrictions that are reversed/lifted within 72 hours and characterized by the company as an error or misunderstanding still count for resolution purposes.

Resolution source: Official statements from OpenAI, xAI, or relevant government agencies (e.g., Commerce Department/BIS), or credible reporting from at least two independent outlets (Reuters, Bloomberg, The Information, etc.) if no official statement is made.

Resolves NO if no such event occurs before January 1, 2027, 00:00 UTC.