Anthropic Fable 5 curbs expose Europe's AI dependency gap
TL;DR
- Authors Igor Sevenard and Richard J. Cook argue Europe's AI Act and Cyber Resilience Act give Brussels rules over frontier AI, but not control of it.
- They cite a June 12 US order suspending foreign nationals' access to Anthropic's Fable 5, an LLM based on the Mythos model, as proof of the gap.
- The piece says closing it needs hundreds of billions of euro in compute, cheap electricity, and data infrastructure, not more regulation.
Europe has spent the past several years positioning itself as the world's rule-writer for artificial intelligence, and by that measure it has delivered. The AI Act and the Cyber Resilience Act give Brussels standard-setting power over how frontier systems behave inside the single market. What a new essay in TechPolicy Press, by Igor Sevenard and Richard J. Cook, argues is that rulemaking has been quietly mistaken for sovereignty, and a recent order out of Washington makes the gap uncomfortable to look at.
The hook is Fable 5, described in the piece as an LLM with guardrails based on Anthropic's Mythos model. On June 12, according to the authors, the US government issued an order suspending access to Fable 5 for foreign nationals. They quote the New York Times' assessment that "the order limiting who can use Anthropic's technology is unusually expansive," reaching even Anthropic's own foreign staff, including, per the essay, "the former OpenAI co-founder Andrej Karpathy."
The argument built from there is straightforward. Regulating a technology, the authors write, "is by no means the same as controlling the capabilities that make it powerful." Europe can decide how a frontier model must behave once it is inside the market, but if Washington can gate who gets to use that model in the first place, the leverage that matters lives elsewhere. The piece points to the EU's own Action Plan on Cybersecurity and AI, which concedes that frontier capabilities are chiefly developed outside the European Union and warns that "without compute, models, and data infrastructure, Europe is bound to remain a vulnerable user."
The material bottlenecks the authors emphasise are the unglamorous ones: cheap continuous electricity, which they call "the absolute baseline of AI development," data centre permitting speed, and talent that keeps leaving for US and Chinese labs. Their proposed price tag for closing the gap is on the order of hundreds of billions of euro.
Take the specifics as reported, not settled. The piece is opinion rather than investigative reporting, and it does not name which EU customers or agencies actually lost Fable 5 access on June 12, nor whether the Action Plan is backed by binding funding. What it does offer is a clear ask for European leaders: treat compute, energy, and permitting as the sovereignty problem, and treat the next order out of Washington as a rehearsal.
Shared on Bluesky by 2 AI experts
Originally reported by techpolicy.press
Read the original article →Original headline: Mythos and the Coming End of European Digital Sovereignty