June was a busy month for AI. The European Parliament gave final approval to the Digital Omnibus, extending key AI Act deadlines while banning certain harmful applications. The CMA gave UK publishers an opt-out from Google’s AI features. And a High Court judge’s criticism of a lawyer for relying on a chatbot prompted fresh debate about AI in legal practice. Meanwhile, the capital kept flowing: SpaceX completed the largest IPO in history before spending $60bn on AI coding start-up Cursor, and Anthropic overtook OpenAI with a $965bn valuation.
What we saw in the UK
The CMA hands publishers an “off-switch” for Google’s AI
The Competition and Markets Authority imposed a conduct requirement on Google’s search business under the new digital markets competition regime, following Google’s designation with strategic market status in general search. The requirement lets publishers opt out of having their content power AI features such as AI Overviews and AI Mode — and opt out of their content being used for fine-tuning Google’s models — while preserving their ordinary search ranking. Google must also attribute publisher content with clear links in AI-generated results and has nine months to comply. With click-through rates falling as users rely on AI summaries, the CMA framed the move as restoring bargaining power to publishers; it is a template other jurisdictions will study.
Britain sets out to build its own sovereign frontier AI model
Announced at London Tech Week on 8 June, UK start-up Cosine unveiled “Lumen Sovereign”, billed as Britain’s first sovereign frontier AI model, co-designed with a coalition of major UK institutions including BT, HSBC, Lloyds, NatWest, LSEG, BAE Systems, Babcock, Leonardo, Thales UK, PwC and the Alan Turing Institute. The model will be trained entirely on Isambard-AI, the Nvidia-powered Bristol supercomputer, using compute awarded under the Government’s £500m Sovereign AI programme. Built on proprietary datasets spanning more than 30 regulated workflows, it is designed to run air-gapped inside a customer’s own infrastructure with no external data transfer, targeting deployment by the end of 2026.
Ofcom publishes its 2026–27 AI strategy
Ofcom’s third annual strategic approach to AI confirms a technology-neutral, outcomes-based stance: organisations in the sectors it regulates may deploy AI without prior approval, provided they meet existing regulatory and consumer-protection duties. The strategy is notable for its enforcement record — Ofcom was the first national regulator to open a formal investigation into a generative AI chatbot (X’s Grok). Deepfakes are a significant focus, informing a draft Fraudulent Advertising Code of Practice scheduled for consultation this summer. The report also identifies agentic AI as an emerging area of regulatory attention.
The ASA warns that existing ad rules apply in full to AI content and deepfakes
The Advertising Standards Authority published guidance confirming that its rules apply in full to advertising regardless of how it is created, including AI-generated content and deepfakes. Advertisers remain responsible for the accuracy and social responsibility of ads produced or manipulated using generative AI, and synthetic endorsements or deepfaked personalities are not a loophole around the CAP Code.
A new data-protection complaints duty takes effect
From 19 June, the Data (Use and Access) Act 2025 requires all organisations to operate a clear data-protection complaints process — acknowledging complaints within 30 days and investigating without undue delay. The ICO has urged SMEs to act, since they are least likely to have formal complaints-handling in place. While not AI-specific, the duty matters for AI deployments: automated decision-making and AI-driven services are precisely where individuals are most likely to complain.
Other UK regulatory signals
Two sector-specific reports deserve note. The Health Research Authority published a 2026–28 plan for AI in health research — a document whose existence tells you how far the technology has moved from lab curiosity to clinical reality. Separately, the Law Commission published its final report on aviation autonomy — the legal plumbing needed before pilotless aircraft can move from trial to scheduled service, covering drones, vertical take-off vehicles, traffic management, passenger safeguards and criminal law.
High Court judge warns against outsourcing legal reasoning to AI
A High Court judge rebuked an associate who had used AI to prepare legal submissions, warning that lawyers must not outsource “the thinking” to a chatbot. The case prompted considerable comment, with practitioners noting the risks of over-reliance on generative AI and the continuing need for professional judgment and verification.
What we saw across the European Union
The Digital Omnibus on AI clears its final hurdle
On 16 June the European Parliament gave final approval — by 423 votes to 57 — to the AI portion of the Digital Omnibus. The amendments push back the AI Act’s toughest timelines: stand-alone high-risk obligations now apply from 2 December 2027 and embedded high-risk systems from 2 August 2028, while watermarking obligations are delayed to 2 December 2026. The legislation also bans AI systems that generate child sexual abuse material or non-consensual intimate imagery (“nudifier” apps), with compliance required by 2 December 2026. The Council must formally adopt the text, and — crucially — most AI Act provisions still begin to apply on 2 August 2026, so businesses should treat the delay as runway, not relief.
Transparency tools land — a final Code of Practice and EU labelling icons
The Commission published the final Code of Practice on the marking and labelling of AI-generated content, drafted by six independent experts with input from more than 180 stakeholders, and released EU icons that deployers may use to label such content. Both support the Article 50 transparency obligations applying from 2 August 2026, which require users to be told when they are interacting with an AI system, machine-readable marking of AI-generated content, and clear labelling of deepfakes. The Code is voluntary but the underlying labelling duties are mandatory, so teams should design disclosures into user journeys now.
The AI Office’s enforcement powers switch on from 2 August
Whatever the Omnibus delays elsewhere, the general-purpose AI regime bites on 2 August 2026. From that date the AI Office can demand technical documentation, training-data summaries and model reports (Article 91); commission independent evaluations including model access (Article 92); and require specific mitigations or withdrawal of a model (Article 93). Article 101 authorises fines of up to 3% of global annual turnover for non-compliance. The AI Office has signalled that technical-compliance dialogues will continue and, if anything, intensify after August, with formal powers used where dialogue proves insufficient.
The Commission unveils a European Technological Sovereignty Package
The Commission presented the European Technological Sovereignty Package, a set of measures to strengthen the EU’s capacity in semiconductors, AI, cloud and open source. Europe relies on non-EU countries for over 80% of key digital products and infrastructure. The package includes a Chips Act 2.0 and a Cloud and AI Development Act (aiming to triple EU data-centre capacity over five to seven years), alongside an EU Open Source Strategy. With US capital dominating this month’s AI headlines, the Commission is framing AI infrastructure as a sovereignty issue.
High-risk classification guidelines go out for consultation
The Commission’s targeted consultation on draft guidelines for classifying high-risk AI systems has been extended to 23 July to give stakeholders additional time to submit their input. Borderline calls are expected in recruitment, workplace monitoring, credit and insurance, educational assessment, public-service triage and AI embedded in regulated products. The practical takeaway is to address classification during product design and procurement, not after deployment.
Poland adopts a draft AI bill
Poland’s parliamentary committee on digitalisation adopted a draft AI bill creating a new national oversight body: the Commission for the Development and Security of Artificial Intelligence (KRiBSI). The body will inspect, penalise and pull non-compliant AI systems from the market, and issue permits for high-risk deployments. Member states are now building the enforcement machinery the AI Act assumes.
Cate Blanchett launches a “Human Consent Registry”
At the European Parliament, Oscar-winning actress Cate Blanchett unveiled a “Human Consent Registry” — a tool letting individuals declare, in machine-readable form, whether and on what conditions AI systems may use their likeness. The initiative responds to disputes over AI models trained on people’s identities without permission. Blanchett praised the AI Act for showing that innovation and rights can coexist, and suggested the registry could become practical infrastructure supporting binding rules. The announcement dovetailed with the Paris Commitment signed by creators’ organisations in the same month.
International developments
The G7 agrees common principles on children’s online safety
G7 digital and technology ministers agreed a common set of principles on children’s online safety, which the Commission welcomed as inspired by the EU’s approach. The principles cover safety-by-design, privacy-preserving age assurance, robust enforcement and better data sharing between platforms, parents and researchers, with strict measures against harmful content — expressly including child sexual abuse material and non-consensual intimate images produced by AI. The agreement is framed around child safety rather than AI specifically, but the AI dimension is explicit and points to growing international convergence around the EU’s model.
IBA launches Artificial Intelligence Institute
The International Bar Association launched its Artificial Intelligence Institute, which will promote responsible AI governance through international cooperation and interdisciplinary expertise. The Institute will draw on the IBA’s global network to address challenges posed by rapidly advancing AI technology.
Business announcements
SpaceX prices the largest IPO in history
SpaceX began trading on Nasdaq under the ticker SPCX on 12 June, pricing at $135 a share the night before, raising about $75bn at a roughly $1.77tn valuation and closing its first day up around 19%. The debut pushed the company past $2tn and more than doubled the previous fundraising record. Proceeds are earmarked for AI compute infrastructure, then launch vehicles and satellite constellations. Days earlier SpaceX had revealed AI1, its first orbital data-centre satellite. The limiting factor for frontier AI is no longer the models themselves, but the physical infrastructure — power, cooling and processing capacity — needed to run them.
SpaceX then spends $60bn on the AI coding start-up Cursor
Four days after its debut, SpaceX announced it would acquire Anysphere, the company behind AI coding agent Cursor, for $60bn in an all-stock deal — the largest ever acquisition of a venture-backed start-up. The arrangement traces back to an April option under which SpaceX secured the right either to buy Anysphere for $60bn or to walk away for an approximately $10bn break-up fee. Folding one of the most widely used AI coding tools into Elon Musk’s xAI ecosystem cements SpaceX’s rapid transformation into an AI powerhouse.
Anthropic and OpenAI head for the public markets
At the end of May, Anthropic announced a $65bn Series H at a $965bn valuation — more than doubling its February mark and, for the first time, clearing OpenAI’s $852bn — and shipped Claude Opus 4.8 the same day. Anthropic disclosed IPO preparations around the start of June, and on 8 June OpenAI followed with its own confidential S-1, setting up two of the most anticipated equity offerings in recent memory. Alongside the S-1, Sam Altman and Jakub Pachocki published “Built to benefit everyone”, declaring a “third phase” for the company built around three goals: an automated AI researcher (with a “significant fraction” of research done by AI by March 2028), faster economic growth, and a “personal AGI for everyone”. Days later OpenAI agreed to acquire Ona, whose customer-controlled cloud environments let its Codex agents keep working inside a company’s own infrastructure — Codex now counts around 5 million weekly users. With private valuations approaching $1tn, the frontier race is now as much a capital-markets contest as a technical one.
Getty puts licensed images into ChatGPT
Getty Images announced a multi-year display partnership with OpenAI to surface its licensed editorial and stock library inside ChatGPT’s search and discovery experiences, with attribution — and, importantly, not to train OpenAI’s models. The deal offers a licensed, “tollbooth” route to visual content for AI-powered search, providing a more authoritative pool of imagery than open-web scraping. It arrives against the backdrop of Getty’s own copyright litigation and its pending merger with Shutterstock. Content owners want to license into AI systems on their own terms rather than litigate after the fact.
Bezos bets on physical AI
Project Prometheus emerged from stealth with a $12bn Series B at a $41bn valuation. Roughly seven months old and employing about 150 people across San Francisco, London and Zurich, it is co-run by Jeff Bezos — his first CEO seat since leaving Amazon in 2021 — and Vik Bajaj, the Google X and Verily veteran. The company is building what Bezos calls an “artificial general engineer”: AI tools that compress the design-to-manufacturing cycle for physical objects, from bridges to chips. Separately, Bezos Expeditions is backing UK start-up CuspAI, based in Cambridge, in a roughly $400m round alongside Kleiner Perkins, a financing that reportedly quadruples the two-year-old company’s valuation to about $2.6bn. Founded in 2024 and advised by Geoffrey Hinton and Yann LeCun, CuspAI uses AI to design new materials with applications spanning AI chips, cooling and clean infrastructure. Both investments point to the same thesis: the next wave of AI value lies in physics-grounded engineering rather than language.
Cohere triples its UK footprint
Canadian AI company Cohere announced it would roughly triple its UK office footprint, moving to 100 New Oxford Street to support R&D growth. The enterprise-focused lab — which in April acquired Germany’s Aleph Alpha to create what it billed as a “transatlantic AI powerhouse” valued at around $20bn — framed the move as an investment in London’s R&D ecosystem and a bet on the UK’s “sovereign AI” market, where government and enterprise demand for secure, private AI is accelerating.
Looking ahead
A theme of June is the widening gap between AI capability and the governance frameworks designed to regulate it. The institutions trying to close that gap are moving at different speeds. From 2 August, the EU’s general-purpose AI rules become enforceable and the AI Office’s powers take effect, even as the UK continues to rely on existing regulators rather than dedicated AI legislation, and the capital markets prepare for the OpenAI and Anthropic listings expected to follow SpaceX.
For businesses, the practical priority is knowing which AI systems are in use, on what terms and subject to what oversight, and being able to show compliance to regulators, courts or clients. The month’s developments around intellectual property and consent — from the CMA’s publisher opt-out to the Paris Commitment and Getty’s licensing deal — show that data flows into and out of AI systems are increasingly a commercial and contractual matter, not just a legal one.
If you would like to chat about any of these developments and what they could mean for your business, do get in touch with Tim Wright or another member of our Technology team. The AI Round-Up now takes a break for the summer and will return in September.
