Policy Implications
Analysis of what the ARC-Eden research programme means for AI policy, governance, and regulation. These posts connect the technical findings to real-world policy questions. 15 posts
- A short reading list for policy readers on AI sovereignty in 2026 What to read after the Chokepoint chapter to understand the policy landscape as of mid-2026. Primary sources, not editorials.
- CADA and the kill switch nobody wants The Cloud and AI Development Act is anti-foreign-kill-switch sovereignty, not a shutdown mandate. Read Virkkunen's quote in context.
- Caretaker Doping and the fourteen-month hardware convergence An April 2025 manuscript concept, then three independent hardware proposals inside fourteen months. How the convergence lines up.
- Chips Act 2.0 without the jargon Demand-side sovereignty, a 120 billion euro target, and why this is not the 2023 Chips Act rebooted.
- Deemed exports and the compute pipeline nobody talks about The rule that treats access as export. What it covers, where it fails, and why cloud-based frontier training raises deemed-export questions the classical framework was not written for.
- Export controls meet cloud APIs: the deemed-export question When a customer overseas queries a US-hosted model, has the model been exported? The Harvard Law Review blog's answer, in plain English.
- FlexHEG explained: the ARIA-commissioned guarantee processor What arXiv:2506.15093 proposes: tamper-proof guarantee processors that make cryptographic claims about what an AI chip is doing.
- From Chokepoint to policy: what a chapter written for readers became in policy documents The Chokepoint argument in Infinite Architects has been quoted, paraphrased and re-derived across five separate policy vehicles in six months. What travelled and what stayed on the page.
- G7 Evian and Macron: the trusted-partners framework in one paragraph What Macron actually said at G7 Evian on 19 June 2026, what the framework proposes, and how it maps to the Chokepoint argument in the book.
- OpenAI plus Broadcom Jalapeno: a nine-month tape-out and no public dollar figure The Jalapeno inference ASIC announcement of June 2026. What is verifiable, what is not, and why the register refuses to invent a headline number the primary sources do not carry.
- Petrie's embedded off-switches, explained without hype What arXiv:2509.07637 actually proposes: thousands of deadman security blocks per accelerator, under 1% die overhead, cryptographic licences.
- Taiwan and criminal liability for chip exports June 2026 proposals extend criminal penalties to unauthorised advanced-chip exports. What the draft says and why it is not just about smuggling.
- The Fable 5 week, day by day 9 June to 1 July 2026: the first export-control action against a deployed commercial AI model, in the order it actually happened.
- The Odeh patents: runtime ethics at the hardware level US 2026/0010411 and US 2026/0010780 propose runtime ethics gates in silicon, with sub-5 microsecond ethical filtering.
- Vertical integration and why the AI stack is collapsing into single owners From chip design through data centre through model. The economic logic driving the collapse, and the safety questions the collapse changes.