Neural Network - June 2026
In this edition of the Neural Network, we look at key AI developments from May and June 2026.
In regulatory and government updates, the European Commission has published draft guidance on the classification of “high-risk” AI systems under the EU AI Act; AI chatbots and social media appear set to face tighter regulation under the UK’s evolving online safety regime; Ofcom has set out its annual strategic approach to safe and secure AI adoption; and the US has introduced a voluntary framework for early government access to AI models.
In technology developments and market news, Anthropic has overtaken OpenAI in the private market valuation race.
More details on each of these developments are set out below.
Regulatory and Government updates
- European Commission publishes draft guidance on classification of high‑risk AI systems
- UK tightens child online safety rules for AI chatbots and social media
- Ofcom sets out strategic approach to safe and secure AI adoption for 2026/27
- US executive order seeks early government access to advanced AI models
Technology Developments and Market News
Regulatory and Government Updates
European Commission publishes draft guidance on classification of high‑risk AI systems
On 19 May 2026, the European Commission (the “Commission”) published its much-anticipated draft guidelines on the classification of high‑risk AI systems under Article 6 of the EU AI Act (the “AI Act”) (the “Draft Guidelines”). The Draft Guidelines are intended to support uniform application of the AI Act by helping providers, deployers and market surveillance authorities determine whether an AI system falls within the high-risk category. Although they are not legally binding, they provide the clearest indication to date of how the Commission expects the classification exercise to work in practice.
The Draft Guidelines are divided into three parts: firstly, addressing general principles that apply across the high-risk categorisation, and then addressing Annex I and Annex III AI systems in turn, as the two applicable routes to high-risk classification. Key points from each section are set out below.
1. General Principles
This first section sets out the general principles relevant to high-risk classification, including the meaning of an AI system and the importance of intended purpose in the classification analysis. The Commission makes clear that intended purpose remains central to any classification and must be assessed by reference to the provider’s instructions for use, technical documentation, promotional materials, and product descriptions. A provider cannot avoid a high-risk classification simply by stating in its terms that high-risk uses are not permitted, if the AI system’s overarching documentation suggests that such high-risk uses are foreseeable. The Draft Guidelines make clear that human involvement does not affect whether a system is classified as high-risk. Instead, human oversight is treated as an essential compliance requirement for systems that are already high-risk.
2. AI Systems in Product Safety
The second section addresses Annex I of the AI Act and the first route to high-risk classification under Article 6(1). This classification applies where: (i) an AI system is itself a regulated product or serves as a safety component of a regulated product (across sectors such as vehicles, machinery, toys, and medical devices); and (ii) the relevant product is subject to a third-party conformity assessment. Both conditions must be met for an AI system to be classified as high-risk.
The Draft Guidelines clarify that the concept of a safety component is broader than might be assumed: it includes not only systems with a direct safety function (i.e. intent-based, where the system’s intended purpose is to prevent or mitigate risks to health, safety or property), but also systems not necessarily intended as safety features but whose failure or malfunction could create or increase health or safety risks (i.e. consequence-based, where a system that helps control how a product operates could still endanger health, safety or property if it fails or malfunctions).
3. AI Systems in Eight Key Areas
The third section addresses the second route to high-risk classification under Article 6(2) of the AI Act, which applies to AI systems intended for one of the eight Annex III use cases. These comprise biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, and the administration of justice. The Draft Guidelines explain the narrow Article 6(3) derogations under which certain Annex III systems may fall outside the high-risk category where they perform only limited procedural, preparatory or similar supporting functions. However, the Commission indicates that these derogations should be strictly interpreted, particularly where a system materially influences decisions affecting individuals’ rights, opportunities, or access to services.
The publication of the Draft Guidelines has come shortly after provisional agreement was reached on the AI Omnibus, under which standalone high-risk obligations are now to apply from 2 December 2027, and obligations for high-risk AI embedded in regulated products from 2 August 2028 (read our article on the EU Digital Omnibus Deal: Simplification of AI Act and Postponed Deadlines). These Draft Guidelines form part of the wider EU shift towards a deferred application of the high-risk regime under the AI Act, giving organisations more time to work through the classification exercise and prepare for future compliance activities.
The Commission has opened a targeted consultation on the Draft Guidelines until 23 June 2026. Although the Draft Guidelines are not yet final, organisations should use them now to assess whether their systems are likely to fall within the high-risk category and to prepare for related compliance work as the Draft Guidelines are finalised.
UK tightens child online safety rules for AI chatbots and social media
The UK Government is continuing to advance its online child safety agenda, with a number of key announcements over the past few weeks.
Following the entry into effect of the Crime and Policing Act 2026 (“CPA”) and the Children’s Wellbeing and Schools Act 2026 (“CWSA”), which both received Royal Assent at the end of April 2026, UK ministers have signalled that chatbot providers and online platforms are likely to face a tougher safeguarding environment - especially where children are concerned – as the UK Government considers further potential legislative interventions to address online harms.
The UK Government is expected to leverage powers granted under these new Acts to amend the Online Safety Act 2023 (“OSA”) via secondary legislation and materially strengthen its ability to regulate AI-generated content. This message has become firmer in recent days, with Sir Keir Starmer warning tech companies that they must do more to address online harms affecting children or face further legislation. Speaking on 8 June 2026, he said that if companies did not act to introduce appropriate protections within three months, the Government would “act and change the law”. We cover this development in more detail in our article: AI Chatbots Facing Tighter Rules under New Laws.
This has been followed by a further significant escalation in the UK's approach to children's online safety, with Sir Keir Starmer’s announcement on 15 June 2026 that social media platforms will be banned from offering services to under-16s. Modelled on similar restrictions introduced in Australia last year, draft regulations are expected before Christmas with the new restrictions due to take effect as early as Spring 2027.
We give our initial thoughts in our article UK Government announces social media ban for under-16s: enforcement and next steps.
Ofcom sets out strategic approach to AI for 2026/27
On 4 June, Ofcom published its 2026/27 strategic approach to AI (the "Report"). The Report has been published in response to the Government's request for regulators to demonstrate how they are supporting the UK's AI growth agenda, and explains how Ofcom applies its technology-neutral, outcomes-based regulatory principles to AI.
The Report sets out how and when Ofcom regulates. Ofcom operates with a “bias against intervention,” aiming to empower its regulated sectors to deploy AI confidently and responsibly, but with a willingness to intervene where required. The Report highlights Ofcom’s proactive monitoring of emerging harms, with Ofcom committing to “promptly and effectively” intervene and exercise its powers where AI-associated risks arise. Ofcom was the first national regulator to launch formal investigations into X’s Grok chatbot and it has committed to continue to act within its powers to address harms exacerbated by AI such as deepfake intimate image abuse.
The case studies illustrate the various tests and investigations Ofcom is conducting, notably by using its own internal data and developing its own AI capabilities. Findings from Ofcom’s Deepfake Defences and Attribution Toolkit papers fed directly into a draft Fraudulent Advertising Code of Practice, with final publication expected this summer.
Looking forward, an annex to the Report details Ofcom’s planned activities over the next year focusing on enabling safe AI adoption and identifying emerging risks. As discussed above, this includes continuing its work to identify emerging and longer-term AI developments that could have implications for citizens and consumers and taking appropriate action to combat non-compliance with the OSA. Ofcom is also preparing for new powers expected to be granted under the CPA and the CWSA, signalling that, while comprehensive AI rules are not imminent, targeted intervention, codes and investigations are arriving quickly.
In light of the under-16s social media ban, the UK Government has asked Ofcom to carry out a rapid study by October 2026 on what effective age assurance looks like for verifying whether a user is over 16. It has also asked Ofcom’s new Chair for an urgent review of the regulator’s enforcement capabilities, with a clear enforcement strategy to follow as soon as possible. The same announcement confirms that Ofcom will be funded to take on these new responsibilities while continuing its existing OSA work on issues such as harmful content, child sexual abuse material, fraud and scams.
US executive order seeks early government access to advanced AI models
On 2 June 2026, President Trump signed an Executive Order titled "Promoting Advanced Artificial Intelligence Innovation and Security" (the "Order"), creating a voluntary framework for the US government to review advanced AI models before they are made publicly available.
Under the Order, AI models that meet a certain capability threshold (referred to as "covered frontier models") are subject to a pre-release review period of up to 30 days. During this window, developers are invited to share their models with the federal government and government-selected partners, subject to confidentiality and intellectual property protections. The capability threshold will be set through a classified evaluation process led by the Director of the National Security Agency, meaning that developers will not know in advance whether their models fall within scope.
The Order states that participation is voluntary and explicitly prohibits any mandatory licensing or government approval requirements for AI development or release. This differs from the Biden Administration's now-revoked Executive Order 14110, which required developers to share technical information about their AI models with the federal government as a matter of legal obligation. That said, the practical incentives to cooperate may be strong, as Anthropic's refusal to grant the Pentagon unrestricted access to its AI models earlier this year resulted in a terminated contract and a national security risk designation.
The Order also addresses AI-related cybersecurity risks. The Cybersecurity and Infrastructure Security Agency are directed to strengthen the security of non-military federal government IT systems, and existing fraud and computer crime laws are to be prioritised for enforcement where AI is used as the means of attack. A new central body will also be established to help government agencies and critical infrastructure operators identify and fix software vulnerabilities.
Technology Developments and Market News
Anthropic overtakes OpenAI with new funding at $900 billion valuation
Anthropic, the US AI company behind Claude, has raised US$65 billion in a new funding round, pushing its valuation to $965 billion. This new funding takes it ahead of OpenAI in the private market as the highest valued AI company globally and comes amid speculation that an initial public offering could follow later this year.
The raise also includes investment backing from chipmakers Micron, Samsung, and SK Hynix, underlining the increasingly close links between AI developers and the infrastructure suppliers they rely on. Anthropic said its annualised revenue crossed $47 billion in May 2026, reflecting the speed at which demand for Claude has grown among enterprise users.
For organisations operating in AI, data protection and cyber security, this is another reminder of the pace at which the market is consolidating around a small number of frontier model developers. It has likely also accelerated Anthropic’s ability to secure the compute capacity needed to support wider deployment of its products, including the 9 June release of Fable 5, a public version of its advanced Mythos model that has already attracted scrutiny from governments and financial regulators. You can read more about Claude Mythos in our previous insight That Mythos Moment: What really happened and why it matters.
The deal highlights both the level of investor interest in frontier AI and the increasing concentration of resources among a small number of leading developers. At the same time, it is likely to sharpen existing questions about the sustainability of current valuations and the increasingly circular relationships between AI developers, cloud providers, chipmakers, and investors that now sit at the centre of the market.
