Our Take
The U.S. just proved what G7 leaders feared: American AI access is revocable without notice, and the stated reason (safety guardrails that exist in rival models) exposes how arbitrary the criteria can be.
Why it matters
Any company or government betting on U.S. AI infrastructure now faces existential supply-chain risk. The Anthropic blackout has made that abstract threat concrete, and it's accelerating pressure on democracies to build independent AI capacity or negotiate carve-outs.
Do this week
Infrastructure teams: audit your production dependencies on Anthropic, OpenAI, or other U.S.-based models this week so you can map fallback routes and cost the risk of sudden unavailability.
Trump administration blocks Anthropic model exports after Amazon tip
The Trump White House ordered Anthropic to halt exports of its Mythos 5 and Fable 5 models on national security grounds. Amazon flagged to the administration that certain safety guardrails in the models could be bypassed. The order came down days before the G7 summit, where the decision became the dominant subtext in AI policy discussion.
At the summit, French President Emmanuel Macron and Indian Prime Minister Narendra Modi publicly objected. Macron warned that if the U.S. "from one day to the next can turn off the switch," it harms not only foreign economies but also the AI firms themselves. Modi raised the same concern, calling unfettered access to top models essential for democratic nations to protect critical infrastructure. Anthropic CEO Dario Amodei was present for the lunch where these objections were aired.
The irony underscores the government's logic problem: cybersecurity experts have documented that the bypass capabilities cited by the White House exist in models that remain freely available, including from OpenAI. No export restriction applies to those competitors. Anthropic's models remain frozen.
A supply-chain chokepoint becomes explicit policy
Until the Anthropic blackout, the risk of U.S. AI export control was theoretical. Now it's operational. Any startup, enterprise, or government that has built a product or workflow on Anthropic's API faces the possibility of loss of access without advance warning and potentially without a stated reason. For mission-critical applications (healthcare, infrastructure, defense), that is unacceptable.
The G7 discussion reveals why democracies are now treating AI sovereignty as a national interest equivalent to semiconductor capacity. Macron proposed a "trusted partners" scheme to bypass U.S. restrictions by allowing model access to nations that agree to use the tech defensively against rivals like China. But even Macron acknowledged the core problem: no one will commit to buying U.S. AI infrastructure if the supply can vanish overnight.
Cohere CEO Aidan Gomez framed it directly: "Companies and democratic nations remaining dependent on a small handful of big tech companies is dangerous to resilience." The Anthropic incident proves the point in real time.
The practical effect is to weaken American AI vendors' negotiating position internationally, even as their models remain technically superior. It also creates urgency around domestic AI development in Europe and India, where lag is otherwise acceptable if independence is the tradeoff.
Audit your AI stack for single points of failure
If your product or infrastructure depends on a single U.S.-based LLM provider, you now have a live use case for contingency planning. Identify which workflows run on Anthropic, OpenAI, or other U.S.-based models. Map the cost and latency of switching to alternative providers (open-source models, EU providers, or multi-vendor setups). Build and test a fallback path before you need it.
For teams in regulated industries or working with government customers, expect more detailed questions about model provenance and jurisdiction. The Anthropic block will be cited as precedent in procurement conversations for the next year.
Open-source model deployment locally becomes a more defensible option, even if the models lag closed-source alternatives. The supply-chain resilience argument now has teeth.