In June 2026, two US decisions showed that access to commercial AI can be delayed or revoked by a single directive. For anyone running AI at the operational core of a trading business, that changes the architecture decision.

Within two weeks in June 2026, the US government did two things it had never done before.
On 12 June, the Commerce Department ordered Anthropic to suspend access to its two most advanced models, Fable 5 and Mythos 5, for any foreign national, whether inside or outside the United States, including Anthropic's own non-citizen employees. Because the company could not screen users by nationality in real time, it disabled both models for every customer worldwide. Days later, after negotiations, access to Mythos 5 was restored for a small group of trusted partners. Fable 5 stayed blocked for general use.

On 25 June, the White House asked OpenAI to limit the release of its next model, GPT-5.6, to a small set of government-approved partners before any wider launch, with access reviewed case by case.
Both are firsts. For years, US controls applied to the chips that train and run AI. The models were outside their scope. This was the first time export controls were used to cut off access to a commercial model, and the first time an administration asked a US company to hold back the launch of a frontier model. Frontier AI is now treated as national security infrastructure, and access to it can be shaped, delayed, prioritised, or revoked.

This is a structural story. The mechanism matters more than the administration. A capability that hundreds of millions of people depended on was withdrawn, globally, with little notice and no detailed public explanation. The same lever exists regardless of who holds it.

It fits a broader pattern. In 2025, a US sanctions decision led to a US provider suspending the email account of the International Criminal Court's chief prosecutor, and later to a US provider disabling a set of services to a unit of a foreign defence ministry. In October 2025, single configuration failures at two of the largest cloud providers cascaded into global outages. Under the US CLOUD Act, a US-headquartered provider can be compelled to produce data it controls, wherever in the world that data physically sits. In June 2026, the European Commission proposed its Cloud and AI Development Act precisely because dependence on a small number of non-EU providers is now treated as a strategic vulnerability. The direction of travel is consistent.

For a commodity trading firm, the exposure is specific. To be useful, AI needs access to contracts, prices, positions, P&L, counterparties and internal workflows. That data is the commercial edge. When the model that reads it sits in a third-party environment outside your jurisdiction, three things stop being in your control: whether the capability is available tomorrow, whether your data is reachable by a foreign legal order and whether a model's behaviour changes under an update you did not ask for.

There is one architecture that removes that dependency: AI that runs where you control it. Euclid deploys on client-owned infrastructure or inside Euclid's private Swiss data centre. The model sits next to the CTRM, inside infrastructure you own. A government directive aimed at a US provider reaches no further. Your trading data stays inside your perimeter. You can see what the system does, because the system is yours.

Today, that private AI layer handles document data integration across contracts, invoices, confirmations and reports, and answers natural-language questions against the CTRM database. The same private AI layer is the foundation for more: the broader operational workflows across Trading, Risk, Operations, Finance and Management that the platform already supports.

Most vendors are adding AI. June 2026 made the more important question plain. The real question is where the AI runs.

See how private AI works inside your CTRM.