Your AI Provider Now Needs Government Permission to Ship. Here's What That Means for Your Workflows.

If you built your workflow around Claude's most powerful model, you cannot use it today. It has been suspended since June 12 under an export control directive, with partial restoration only for US organisations that operate critical infrastructure. If you were waiting for GPT-5.6 Sol — OpenAI's new flagship — you are behind a government approval process that currently covers roughly 20 companies. For the first time, both leading AI providers have their best models gated not by a pricing tier or a waitlist, but by a government decision.

What happened this week

OpenAI launched GPT-5.6 in three tiers — Sol (flagship), Terra (balanced), Luna (fast) — but only as a limited preview for approximately 20 government-approved companies (June 26). OpenAI said publicly that "we don't believe this kind of government access process should become the long-term default," which suggests they did not find it optional. Anthropic partially restored Mythos 5 access (June 27) but only for US organisations that "operate and defend critical infrastructure." Fable 5, the consumer model, has been suspended since June 12.

The clearest signal that the disruption is real comes from the companies themselves: OpenAI, Anthropic, Microsoft, Amazon, Bank of America, GM, IBM, and Eli Lilly jointly backed RAISE US (June 25), a bipartisan nonprofit with more than $500 million to retrain AI-displaced workers, piloting wage insurance and apprenticeships in four states. Companies do not fund $500 million retraining programmes for disruptions they consider manageable.

Three things to do differently this week

Test your backup. If your primary AI tool disappeared today, could your work continue? Fable 5 users have been locked out for more than two weeks. GPT-5.6 users outside the approved group have no firm timeline. This is no longer a theoretical risk — it already happened, twice, to the two most widely used professional AI tools. Identify which tools your current workflows depend on, and test at least one alternative for each critical step.

Build workflows that are not tied to one provider. Processes built around a specific model's quirks, reasoning depth, or API behaviour will break when that model changes, retires, or becomes restricted. Processes built around clear inputs, specific output criteria, and structured prompts tend to transfer. The distinction matters more now than it did three weeks ago.

Invest in the judgment skills that no government order can restrict. The companies building AI are spending half a billion dollars on retraining because they know which capabilities AI will absorb and which it will not. Evaluation, knowing when to override an output, and defining what good looks like before the work starts — these are not model features. They are professional skills, and they are worth building now.

Maria: three days lost, four hours to fix

Maria is a content strategist at a 55-person B2B software company. Her core production workflow — brief to first draft to SEO review — ran entirely through Claude. On June 12, Fable 5 was suspended. Her team spent three days testing alternatives before landing on a working substitute. The content calendar slipped by a week.

She now keeps a tested backup workflow: the same prompt templates, validated against a second provider, updated quarterly. The backup took four hours to set up. The disruption cost three days. She will not make that trade again.

Ryan: thirty minutes, three gaps

Ryan is an operations manager at a 200-person logistics company. His team uses AI to draft carrier communications, track exception handling, and summarize daily operations reports. When he read about the RAISE US fund, the size of it — $500 million, backed by the same companies that build the AI — struck him as a signal worth acting on.

He ran a 30-minute audit: which parts of his team's work could not continue if their primary AI tool went offline for a week? The answer covered more ground than he expected. Three core workflows had no manual fallback and no tested alternative. He now has one workflow per month on the roadmap until each has a viable backup, and a clear sense of which steps in each workflow actually need AI and which ones just migrated there by habit.

The one-sentence version

Both leading AI providers now have their best models behind a door you do not control — test your backup, build workflows that are not tied to a single provider, and invest in the judgment skills that no restriction can take from you.