Omnibus weekend read. Europe Day. NHS GitHub goes private today.

On 8 May 2026, the European Commission published draft guidelines on the implementation of the transparency obligations under Article 50 of the AI Act, opening a targeted consultation that runs until 3 June 2026. The guidelines are non-binding, but they are the first Commission instrument to provide interpretive guidance across the full scope of Article 50, prepared in parallel with the second draft of the Code of Practice on Transparency of AI-Generated Content published on 5 March. On Tuesday 12 May, the international law firm Covington & Burling published a detailed ten-takeaway analysis on its Inside Global Tech blog — the first substantive industry legal review of what the document actually means in operational practice.

The substance is significant. From 2 August 2026, providers of AI systems intended to interact with natural persons must inform users that they are interacting with AI — unless this is “obvious from the point of view of a reasonably well-informed natural person, taking into account the circumstances and the context of use.” Providers of generative AI must implement machine-readable marking of synthetic audio, image, video and text content. Deployers must disclose deepfakes, AI-generated public-interest publications, and the use of emotion recognition or biometric categorisation systems. The Omnibus deal struck on 7 May grants a transitional period until 2 December 2026 for systems already on the EU market before 2 August; systems placed on the market from 2 August onwards must comply from that date.

The Covington analysis identifies the practical pressure points. The Code of Practice addresses only the marking obligations under Article 50(2) and the deepfake-labelling obligations under Article 50(4) — the Guidelines cover everything else. Operators of conversational systems will need to determine what counts as “obvious” AI interaction in their context. Deployers of generative AI will need to operationalise machine-readable marking that survives downstream editing. Companies will need to map their AI inventory against the four distinct transparency obligation streams (50(1), 50(2), 50(3), 50(4)). With under twelve weeks to the deadline, no certified standards, and a Code of Practice not finalised until June, the operational uncertainty is real. For European companies, the practical effect is that the transparency layer of the AI Act is now the layer arriving fastest. The high-risk Annex III deadline has moved to 2 December 2027. Transparency is operational this summer.

Today, 13 May, is the deadline for interested parties to submit feedback on the European Commission’s draft measures in case DMA.100220 — the specification proceedings on Google Android interoperability with third-party AI services under the Digital Markets Act. The Commission opened proceedings on 27 January 2026. Preliminary findings were addressed to Alphabet on 27 April. The Commission must adopt its final binding decision by 27 July 2026 — within six months of opening the proceedings.

The substance is operationally significant. The Commission’s draft measures cover four main themes. Wake words: third-party AI apps would be able to be invoked by a custom voice phrase at any time, even when the screen is locked, with minimal battery impact. System-wide access points: long-pressing the home button or navigation handle would invoke third-party AI services, with those services receiving contextual data to enable functions like translating on-screen text or searching for information on screen — capabilities currently reserved for Google’s Circle to Search. Effective interaction with apps: third-party AI services would be able to send emails using the user’s preferred email app, order food, or share photos — the agentic functions Google reserves for Gemini today. Hardware and software resources: access to background execution, memory, and other system resources to ensure reliability and responsiveness.

Executive Vice-President for a Clean, Just and Competitive Transition Teresa Ribera framed the case directly when the preliminary findings were published in April: “AI services are becoming more and more relevant for EU citizens’ daily interaction with their mobile devices.” The case is, in practice, the single most consequential AI-platform competition file in Europe right now. If adopted as drafted, the measures would materially restructure the competitive position of every non-Google AI assistant operating in the EU smartphone market — particularly Mistral’s Le Chat, OpenAI’s ChatGPT, and Anthropic’s Claude. Final binding measures take effect on 27 July. The clock is now running in days, not months.

According to Crunchbase data published Tuesday, roughly half of European venture funding in 2026 to date has been in AI-related companies. Total European startup funding reached over $17 billion in Q4 2025 and again in Q1 2026 — roughly a third higher year-on-year. The most striking concentration is at the frontier-lab tier. Three new labs founded by senior researchers departing Big Tech have, together, raised more than $2.6 billion in 2026 alone, anchored in two European cities.

Ineffable Intelligence, founded in London by former Google DeepMind principal scientist David Silver, raised $1.1 billion in seed funding at a $5.1 billion valuation in late April. Backers include Sequoia, Lightspeed, Nvidia, and Google. Silver’s pitch is reinforcement learning at frontier scale — AI systems that learn from experience rather than human data, training a “superlearner” that develops knowledge through self-play and direct interaction with environments. UK Science and Technology Secretary Liz Kendall framed the round directly: it underlines the UK’s determination “to ensure that the UK isn’t just an AI taker but an AI maker.”

Advanced Machine Intelligence Labs, founded in Paris by former Meta Chief AI Scientist and ACM Turing Award winner Yann LeCun, raised $1.03 billion in March 2026 at a $3.5 billion pre-money valuation. Backers include Bezos Expeditions, Nvidia, and Samsung Electronics. CEO Alexandre LeBrun (founder of Paris health-tech Nabla, former Facebook AI research engineer). AMI’s pitch: world models — AI systems that understand the physical world, maintain long-term memory, and strategise complicated tasks. LeCun’s framing in Reuters: current LLM approaches based on predicting the next word or pixel will not produce broadly capable intelligent agents.

Recursive Superintelligence, founded in London by former DeepMind principal scientist Tim Rocktäschel, was reported in late April to be closing a round of approximately $500 million with capacity for up to $1 billion. The pattern across all three is consistent: founders from London and Paris, technical missions distinct from the dominant LLM paradigm, and investor bases dominated by US venture capital, Big Tech corporate investors, and Asian strategic backers. The labs are European in their geography and their founding teams. The capital tables are not. For European policymakers grappling with the AI Continent Action Plan and the gigafactory programme, the funding pattern of 2026 is a structural data point: Europe can incubate frontier-lab founders. The capital architecture that sits around them is still substantially being built from outside.

The Commission’s digital strategy update on Tuesday confirmed a continuing pattern in European public opinion: 92% of Europeans view the need to further strengthen children and young people’s protection online as a top policy priority, with 93% concerned about social media’s mental health impact and 92% supporting effective restrictions on access to age-inappropriate content. The 2025 Eurobarometer figures continue to anchor the political mandate behind the European Commission’s child-safety legislative pipeline through 2026.

The Omnibus deal reached on 7 May added a new Article 5 prohibition to the AI Act, banning AI systems used to generate child sexual abuse material or non-consensual intimate imagery — the so-called nudifier ban. The prohibition applies in three configurations: placing such systems on the EU market, placing them on the market without reasonable safety measures, and deployer use. Companies have until 2 December 2026 to bring affected systems into compliance. Co-rapporteur Michael McNamara (Renew Europe, Ireland), who carried the file through Parliament’s Civil Liberties Committee, described non-consensual intimate imagery as “a systemic harm being industrialised by AI” that falls overwhelmingly on women and girls.

The institutional pipeline beyond the nudifier ban is dense. President von der Leyen’s Special Panel on Child Online Safety, convened on 5 March 2026 and met again on 16 April, will hold its third meeting in June and deliver final findings to the President by summer 2026. The EU age-verification app, blueprinted on 14 July 2025 and feature-ready as of 15 April 2026, is being adopted by Member States ahead of legislative mandates. The Commission has adopted a recommendation urging Member States to accelerate rollout by the end of the year. France has approved a 15-year social media age limit. Spain, Austria, Greece, Ireland, Denmark, and the Netherlands are preparing parallel measures. The Omnibus nudifier ban is the only part of the deal that civil society organisations including AlgorithmWatch did not characterise as a concession. The political mandate behind it is also the only part of the AI policy agenda where 92% public alignment puts substantive constraint on legislative drift.