How would you like to read it?
Same edition, explained without the jargon — and just as faithful. It's not a quick summary: an independent check confirms the plain-language version stays true to the original, without dropping or distorting anything.
European Commission Adopts the Proposed Cloud and AI Development Act (CADA)
The European Commission has adopted the proposed Cloud and AI Development Act, which focuses on research and innovation, data center capacity, and a single European framework for cloud and AI sovereignty.
On June 3, 2026 the European Commission adopted the proposed Cloud and AI Development Act (CADA), an initiative designed to strengthen the Union's cloud and AI ecosystem, investment, and infrastructure. The measure is structured around three pillars. The first is research, development, and innovation, supporting the rollout of next-generation, sustainable cloud and AI technologies. The second is capacity: accelerating the deployment of data centers across the EU, with particular attention to facilities that strengthen public sector functions. The third is autonomy, through the creation of a single European-level assessment framework for cloud and AI sovereignty, paired with a dedicated adoption mechanism for the public sector.
On the infrastructure side, the Commission is betting on the deployment of AI factories and gigafactories, designed to provide European businesses and researchers with next-generation, high-capacity computing resources. The proposal is complementary to other initiatives already underway, including the Apply AI strategy, the Chips Act 2.0, and the EU open source strategy.
In terms of numbers, the official EU pages set the goal of at least tripling the EU's data center capacity over the next 5-7 years and fully meeting the needs of businesses and public administrations by 2035. The proposal's library page, however, does not report specific investment amounts in its summary. On the sovereignty front, CADA does not provide for a blanket ban on non-European vendors, but introduces sovereignty levels for the public sector and critical cases. These levels include EU data localization, independence from third countries and, for the higher levels, EU ownership and control, with possible recognition of third-country providers by the Commission.
Why it matters
- Entrepreneurs: CADA mobilizes the European cloud and AI ecosystem with AI factories and gigafactories: for businesses, this means potential access to advanced computing resources and a new competitive arena, especially toward public sector demand.
- ICT engineers / IT managers: The single sovereignty assessment framework and the public sector adoption mechanism will shape procurement criteria and architectural choices. The sovereignty levels set out for the public sector and critical cases cover EU data localization, independence from third countries and, at the higher levels, EU ownership and control. The push to triple data center capacity within 5-7 years directly affects infrastructure planning.
Anthropic Extends Project Glasswing to Around 150 Critical Infrastructure Organizations
Anthropic's software security program puts advanced AI models in the hands of those who run critical infrastructure: more than 10,000 serious vulnerabilities already identified by the first partners.
On June 2, 2026, Anthropic announced the expansion of Project Glasswing, its software security program, to around 150 new organizations. The project puts advanced AI models in the hands of those who run critical infrastructure and widely deployed codebases, with the goal of finding and fixing vulnerabilities before they are exploited. After an initial phase that began in early April with around 50 partners, the network now covers organizations in more than 15 countries, working across the energy, water, healthcare, communications, and hardware sectors. Many of the new partners are vendors whose codebases support governments and large populations.
The technical core of the program is the Claude Mythos Preview model. According to Anthropic, the first partners have already identified more than 10,000 high- or critical-severity vulnerabilities. The company estimates that a major attack against most of the codebases involved "could affect more than 100 million people," with consequences for national and global security. The stated rationale is the near-term prospect of AI models that are "cheap and fast with powerful offensive cyber capabilities."
In parallel, Anthropic introduced Claude Security, which uses Claude Opus 4.8 to scan codebases. It has also made vulnerability-finding tools available on request to trusted security teams. Future plans include extending the program to other infrastructure providers and open source project maintainers, as well as strengthening the Cyber Verification Program.
Why it matters
- ICT engineers / IT managers: For those who manage systems and security, the program promises an accelerated stream of patches for serious CVEs in everyday components: the more than 10,000 high/critical flaws already surfaced signal that many critical dependencies and codebases contain problems that have not yet been fixed. It is worth monitoring advisories from the vendors involved and prioritizing timely patching.
OpenAI Codex Adds Sites and Six Vertical Plugins for Office Work
With its June 2, 2026 announcement, Codex moves beyond pure programming: it publishes web apps shareable via URL and introduces six plugins designed for professional roles, from data analysis to investment banking.
OpenAI has extended Codex beyond programming, turning it into a platform for office work. The official announcement on June 2, 2026, accompanied by the report "The Next Era of Knowledge Work", introduces three new features.
The first is Sites: Codex no longer produces only local files, but can publish its output as interactive web apps and pages, hosted and shareable via URL within the workspace. A financial model thus becomes a scenario simulator, a launch plan an updatable hub. For this feature OpenAI has partnered with Wix, Base44, Replit, Lovable, Figma, and Emergent. The second is Annotations, which lets you select specific portions of a document to direct targeted commands and context at them.
The third, and the most significant for businesses, is a family of six vertical plugins designed for specific roles: data analysis, creative production, sales, product design, equity investing, and investment banking. Each plugin packages integrations, instructions, and context to approximate a specific job function. Altogether, the six plugins aggregate 62 applications — including Snowflake, Databricks, Salesforce, and HubSpot — and 110 preconfigured skills.
The move is driven by usage numbers: according to TechCrunch, Codex has surpassed 5 million weekly active users (six times the level at the February 2026 launch of the desktop app). Knowledge workers, currently about 20% of the base, are growing at triple the average rate. "AI is becoming capable of doing increasingly meaningful work inside organizations," said Denise Dresser, Chief Revenue Officer at OpenAI.
Why it matters
- End users: Those who don't write code — financial analysts, salespeople, designers, marketers — get agents already tailored to their role and can turn Codex's work into web apps shareable with a simple link, without going through IT. Codex stops being a tool for developers alone and becomes a surface for everyday productivity.
- LLM builders / devs: Codex's center of gravity shifts from writing code to orchestrating agents across an ecosystem of 62 apps and 110 skills: for those building with LLMs, this means new integration surfaces (Sites as hosting for generated apps) and direct competition in the knowledge work layer, where Anthropic, Microsoft, and Google are also moving.
An OpenAI General Reasoning Model Disproves Erdős's Unit Distance Conjecture
A general-purpose OpenAI reasoning model produced a counterexample to a discrete geometry problem open since 1946. The point isn't that it invented new mathematics: it connected existing tools from algebraic number theory — never before applied to this problem — extending Erdős's own construction. A group of mathematicians verified it and rewrote it by hand.
OpenAI has announced that one of its reasoning models has disproved Erdős's unit distance conjecture, a discrete geometry problem open since 1946. The question, posed by Paul Erdős, is elementary to state: given n points in the plane, how many pairs can lie at a distance exactly equal to 1? For decades it was believed that, with many points, a square grid was close to the optimal arrangement — the maximum possible number of unit-distance pairs.
The model showed that one can do better: it constructed an infinite family of configurations with more unit-distance pairs than the conjecture allowed, with a gain of polynomial order over the expected bound. But the how matters more than the result. The model did not invent a new technique: it took Erdős's own construction and extended it by importing tools from an area far from combinatorial geometry — algebraic number theory, and in particular class field towers (ideas attributable to Golod–Shafarevich, Ellenberg–Venkatesh and Hajir–Maire–Ramakrishna) — that no one had thought to apply to this problem.
This is the nuance that matters. The model's strength was not to "solve" from scratch, but to connect pieces of existing mathematics living in different fields: exactly what a system trained on vast amounts of literature can do better than any single human. According to OpenAI the proof came from a general-purpose reasoning model, not trained specifically for mathematics nor aimed at this problem. A group of nine mathematicians — including Noga Alon, W. T. Gowers and Thomas Bloom — verified the counterexample and published a short, hand-checked version on arXiv. It's not "AI doing mathematics in our place," but AI helping us connect tools we already had.
Why it matters
- Frontier research: A general-purpose reasoning model contributes to an open, central problem in mathematics not by inventing a new technique, but by connecting combinatorial geometry and algebraic number theory — two distant areas — with existing tools never applied here. It moves AI toward the role of collaborator in discovery, and makes human verification of the result a decisive step.
- LLM builders / devs: The technical point is that no specialized system was required: no fine-tuning on mathematics, no scaffolding for proof search, no targeting of the individual problem. The value that emerged is the ability to retrieve and recombine existing tools from different fields — a signal for how to design and evaluate models, rather than "discovery from scratch".