CADA
THE CLOUD AND AI DEVELOPMENT ACT — EUROPE'S TECH-SOVEREIGNTY PLAY DROPPED THIS WEEK
The European Commission proposed a regulation that treats cloud and AI as sovereign infrastructure. OpenAI responded the same week with three policy posts, a democratic-governance blueprint, and a Bedrock partnership. None of those were on the calendar a month ago.
The week the EU told the US tech industry its cloud was sovereign infrastructure
The European Commission proposed the Cloud and AI Development Act on Thursday. Read alongside the wider tech-sovereignty package the Commission released the same day, the message is the one the US tech industry has been quietly bracing for since last summer: critical compute is treated, from now on, as sovereign infrastructure. The legislation does not name OpenAI, AWS, Google, or Microsoft. The legislation does not have to.
OpenAI's week, on the same calendar: a piece on Tuesday titled "Our views on AI policy and political advocacy." A Wednesday case study on youth safety in EMEA. A Thursday "Blueprint for democratic governance of frontier AI" and a separate "OpenAI public policy agenda." A Friday announcement that OpenAI frontier models and Codex are now available on AWS. Five policy and infrastructure pieces in five days. None of them are wrong, individually. Together, they are a posture.
The posture is: we are the responsible adult in the room, we welcome regulation, our models will be on every cloud you need them on, and our blueprint for democratic governance is already written so the Commission does not have to write it for us. The honest read of CADA is that the Commission is going to write it anyway, and a blueprint written by the regulated entity carries the weight blueprints written by regulated entities have always carried.
Two technical signals worth noting under the policy noise: Bedrock Claude Sonnet 4.5 missed our citation-pubmed-v1 probe at -5.29σ twice this week, including on Saturday, which is unusual for a model that is reliably retrained mid-week. Haiku 4.5 missed three times. The drift we first surfaced in W19 is now stable across two providers and into a third week. If you were waiting to see whether the drift would self-correct, the answer this week is: not yet.
A blueprint for democratic governance of frontier AI — written by the largest builder of frontier AI, published the same week the regulator proposes the act it expects the blueprint to influence. The strategy is the timing.
Want to spot this in your own conversations?
CLEAR is the free six-lesson course on the patterns AI quietly runs on you.
Take the course →
Founder's note — Quieter wire-count this week (17), heavier signal. The provider-side noise the new source filter dropped tomorrow's corpus run by about a third. If next issue lands lighter, that's the reason — same signal, less filler.
◆The Notebook
A new EU regulation treating cloud and AI as sovereign infrastructure. Part of a wider tech-sovereignty package. Worth reading the proposed compliance timeline before assuming this is a 2028 problem.
via EU AI Office
Bedrock Claude Sonnet 4.5 missed at -5.29σ twice this week. Haiku 4.5 missed three times. The drift we surfaced in W19 is now stable through W21. Two providers, three weeks, one probe. A reader running an internal version of the same probe sent us a chart that matches.
via AI Newswire probe log
OpenAI published five policy / infrastructure posts in the five business days after the EU tech-sovereignty package, including a blueprint for democratic governance and a public-policy agenda. The cadence is the story; nothing in the calendar a month ago suggested this week.
via OpenAI blog
◆Worth Your Time
EU AI Office
Read alongside the wider tech-sovereignty package. The data-residency provisions are the part that matters; the AI provisions follow.
OpenAI
A well-argued document. Worth reading. Worth reading it in the calendar position it was published in.
OpenAI
Stripped of rhetoric, this is the company's ask of regulators. The asks are reasonable. They are also asks.
EU AI Office
The umbrella the CADA proposal sits under. Read this first, the CADA proposal second.
OpenAI
The multi-cloud play. Notable for the timing rather than the content — the post landed two days after CADA dropped.
The Probe · Test Yourself
The EU AI Act's serious-incident reporting requirement kicks in on 2 August 2026. Your company uses a frontier LLM in a customer-support workflow. Which event would oblige you to file a serious-incident report, under the Act's definition?
AThe model produced a confidently wrong response to a single customer
BA customer relied on a model response and suffered serious harm to their health
CThe model refused a routine query for a week
DInternal latency on the model spiked by 300%
Reveal the answer
Answer: B — A customer relied on a model response and suffered serious harm to their health
Article 73 triggers a reporting obligation when an AI system "directly or indirectly leads to" — among other things — serious harm to a person's health, death, critical-infrastructure disruption, or infringement of fundamental rights. A single wrong response is not enough; harm is the trigger. Latency and refusal are operational issues, not Article-73 incidents. Pin this rubric to the wall before August.
Reply and tell me what you've noticed. If you're building Article 73 incident-reporting workflows for August, send me what you've drafted — I'm collecting a small corpus to share back with subscribers in late June.
Free where it can be. Honest where it has to be.
— Three places to go from here —
Course
CLEAR
Six free lessons on the patterns AI runs on you.
Start →
Tool
LiveScope
Chrome extension that flags what AI cites without checking.
Install →
Read
The Agreement Trap
15-chapter book on living inside the exchange. £5.99 lifetime.
Read →