When we wrote about the Fable 5 shutdown on 13 June, we expected a brief stand-off and a quick negotiated return. Twelve days later, both Fable 5 and Mythos 5 are still dark for every customer on earth. Time to revisit what we got right, what we missed, and what it means now.
What's held up
The framing we used - that this was a policy decision rather than a model failure - has only gotten stronger. Anthropic's senior engineers have been in Washington running what sources have called a crisis negotiation. Reuters has now seen Commerce Secretary Lutnick's letter to Dario Amodei; it cites the Export Control Reform Act of 2018 and names China and Russia as the underlying concern. President Trump told reporters at the G7 that negotiations were "going fine." More than 80 cybersecurity executives, including leaders at Nvidia and Adobe, signed an open letter backing Anthropic's technical position.
That doesn't look like a story about a faulty model. It looks like a story about who gets to decide who uses frontier AI.
The vendor concentration point also held up. On 23 June, Anthropic quietly removed Fable 5 from its Pro, Max, Team and Enterprise plans. Customers paying for those tiers no longer have access through their subscription, regardless of how the negotiation ends. If your business was relying on Fable 5 specifically, you've now had nearly two weeks to think about Plan B — and if you haven't started, that's the thing to do today.
What we underestimated
The duration. We expected a deal in days. We're now at twelve, and prediction markets price only 57% odds of restoration before 1 July.
The framework angle. Trump signed an executive order on 9 June — the same day Fable 5 launched — setting up a voluntary pre-release access framework for frontier models, with a 1 August deadline for the NSA, Treasury and CISA to design it. Analysts at Andrew.ooo and ExplainX have argued the export control directive isn't really a response to the jailbreak claim. It's leverage to force Anthropic into that framework. If that reading is right, this is bigger than one model. It's the first real test of how frontier AI gets governed at a national level, and every major provider is watching how it resolves.
The geopolitical knock-on. President Macron reportedly warned at the G7 that no country will buy American AI if it can be switched off without warning. Canadian PM Carney said the episode shows why countries need to diversify their AI supply. Those are not throwaway lines from second-tier officials. They reframe the question for every government and every business that buys software from US vendors - including Australian businesses that depend on US-hosted AI services.
What to do this week
Our original advice stands, but twelve days of evidence makes the case more bluntly. If you're using Claude in production, have a tested fallback already running on GPT-5.5 or Claude Opus 4.8. Not "we'll figure it out if something happens." Actually running, with a current eval.
The broader point is this: the AI tools your business relies on sit on supply chains and regulatory frameworks you don't control, and those frameworks are being actively contested right now. The Fable 5 episode looked like a narrow, temporary disruption on day one. On day twelve it looks like an opening move in a longer argument about AI governance that US, European and Australian regulators are all watching closely.
We'll write again when there's a resolution. At this rate, that may not be soon.
Sources: Reuters via The Globe and Mail; Anthropic statement on the US government directive (anthropic.com); ExplainX, "Is Fable 5 Back? Status & Alternatives" (June 24, 2026); TechTimes, "Fable 5 Export Ban Day Six"; Andrew.ooo, "What the US Export Control Order Means"; chorrocks.substack.com, "What Happened to Anthropic's Fable 5: The Facts, Part II"
