• The Velocity White Papers is an interactive publication from the Government of Alberta setting out how it is using AI to rapidly and securely modernise its technical estate — the same programme behind the Anthropic/Claude cybersecurity case study covered in the previous batch.

• The UN’s Independent International Scientific Panel on AI released its first Preliminary Report on 1 July 2026, a global, independent scientific assessment of AI’s opportunities, risks and impacts whose central message is that current safeguards are not keeping pace with AI’s advancing capabilities.

• The ONS’s Artificial intelligence in UK businesses: 2023 to 2026, finds that self-reported AI use among firms with 10+ employees has risen from around 12% to 35% since late 2023, though adoption remains shallow — averaging 1.6 technologies per firm — and has so far had little effect on overall workforce headcount.

• UK Thematic Review and Gap Analysis on AI Security, conducted by Lancaster University, maps 9,109 peer-reviewed papers into 12 themes and identifies five priority research gaps, including data integrity, provenance of third-party models, and the security of emerging agentic-AI systems.

• The Digital Identity Sectoral Analysis Report 2026, estimating 275 UK firms generating around £2 billion in annual revenue, with rising productivity (GVA up 17%) alongside a modest fall in employment linked to consolidation and automation.

• DSIT’s The Economic Impact of Equal Opportunities in the Digital Economy models the potential gains from removing barriers facing under-represented groups in the digital sector, estimating up to £16 billion a year from addressing women’s under-representation and £15 billion from that of workers aged 50–69.

• UK Public Engagement Survey 2025/2026 reports that 97% of UK adults are aware of AI and 59% have used generative AI in the past three months, while noting widespread concern about accuracy (74%) and strong public support for AI legislation.

• The EU AI Office’s report draws on over 100 experts, who conclude that the next one to two years are decisive for Europe and identify computing infrastructure and the energy to power it as the most urgent priorities.

• The OECD’s Skills in the AI Age examines how AI is reshaping labour markets and skill requirements, and sets out the policy responses needed to help workers adapt while managing the risk of displacement.

• The Global Index on Responsible AI 2026 (2nd Edition) assesses 135 countries and finds that responsible-AI governance is expanding but not fast or evenly enough, with a persistent implementation gap — frameworks show evidence of implementation in only 55% of cases globally, and fewer than one in five governments disclose their own use of AI.

• The Alan Turing Institute’s AI Governance around the World: China, profiles China’s application-specific regulatory approach — covering recommendation algorithms, deep synthesis, generative AI and, newly, anthropomorphic interaction services — alongside its centrally coordinated standardisation system and growing engagement in international standards.

• The Knowledge Centre Data & Society’s From Policy to Practice: Prototyping the EU AI Act’s Value Chain Requirement, tests the written-agreement obligation under Article 25(4) for high-risk AI systems, finding the provision’s broadness leaves too much room for divergent interpretation and calling for clearer guidance, particularly for SMEs.

• Stanford HAI’s The Commercial Landscape of AI Sovereignty Offerings , surveys sovereignty products across the AI stack and finds that most reconfigure rather than eliminate dependence on a few upstream providers, arguing the real challenge is calibrating interdependence, not avoiding it.

• New Tools for Mapping a Fragmented AI Governance Landscape reports on a dialogue around the “Decoding AI Governance” report, which proposes three tools — a governance stack, a map, and an “Anchors and Hooks” framework — to bring coherence to a sprawling AI governance landscape without forcing full convergence.

• Between Simplification and Stagnation: The EU Digital Omnibus and the Future of Data Protection Law examines the EU Digital Omnibus and its proposed GDPR amendments, arguing that the central question is not whether EU data protection law should be reformed but how it can address persistent structural weaknesses without eroding its legitimacy.
• Google’s AI & Economy ATLAS v1.0: Mapping Gemini Usage in the Economy — Google’s first ATLAS report maps around 15 million de-identified Gemini interactions across 800 occupations and 4,000 tasks, finding that most workplace use supports collaboration such as ideation, strategy, and learning, while full task automation appears in under 10% of interactions.
• Twin Transition: To the Digital, to the Green, or to Somewhere in Between? — This editorial introduces a European Journal of Risk Regulation special issue on the EU’s “twin transition,” asking whether the digital and green agendas are genuinely paired or in practice follow different regulatory paths.
• Design Practitioners’ Perspectives on Digital Health Implementation in Complex Systems interviews with design professionals find little consensus on when a digital health service counts as “implemented,” leading the authors to propose treating design as a contributor to the implementation phase rather than a separate stage that precedes it.
• Triple Helix: Markets, Fundamental Rights and Security in EU Digital Law argues that security increasingly predominates in EU digital law but does not displace the internal market or fundamental rights, offering a framework in which rights safeguards provide only a limited check on creeping securitisation.
• Regulation in the Making: The Governance Effects of Civil Liability in EU Digital Regulation argues that civil liability in EU digital regulation does more than enforce duties or compensate harm, since the structure of compensation claims also specifies regulatory duties and reallocates regulatory risk among market actors.
• AI as Creative Partner: Exploring Perceived Roles in Human–AI Co-creation interviews with 19 architecture students using generative AI tools identify 18 roles that humans and AI take during co-creation, alongside tool-specific variations and insights into emotional dynamics, agency, control, and authorship.
• Regulatory Autonomy and AI Sovereignty: Türkiye at a Crossroad — The authors use Türkiye as a case study of how a “middle power” pursues AI sovereignty, arguing it asserts regulatory autonomy by balancing innovation and control while remaining constrained by limited command over critical infrastructure such as advanced chips.
• The Case for Globally Beneficial Technology — The authors argue that advanced technologies such as AI should be designed, developed, and distributed to benefit everyone, grounding the claim in five moral arguments spanning human rights, beneficence, contingencies of birth, the global tree of knowledge, and global economic justice.
• Do LLMs Really Need 10+ Thoughts for “Find the Time 1000 Days Later”? Towards Structural Understanding of LLM Overthinking — The authors find that long-reasoning models are five to twenty times slower on simple tasks with no real accuracy gain, and introduce an analyser called TRACE that traces their thought process, identifying over-verification and over-exploration as the main drivers of “overthinking.
• The Fragility of AI Companionship: Ontological, Structural, and Normative Uncertainty in Human–AI Relationships — Drawing on interviews with 25 users of AI companions, Zhang and Xie identify three forms of uncertainty in these relationships — about the AI’s nature, about platform control and system instability, and about the legitimacy of human–AI intimacy — and note that users manage them through information seeking, expectation adjustment, avoidance, and disengagement.
• The Authenticity Gap: Why Generative AI Cannot Replicate Human Creativity — The authors evaluate generative AI across originality, effectiveness, aesthetics, and authenticity, finding that human and AI creativity converge on the first two and partly on aesthetics but diverge on authenticity, which they argue requires autobiographical grounding and self-integration that current AI lacks, and so position generative AI as a collaborator rather than a replacement.
• The Dark Side of Artificial Autonomy: Formation and Coping of Relationship-related Technostress between Humans and AI — Through a two-wave panel survey, the authors find that an AI agent’s autonomy has non-linear effects on relationship-related technostress, forming an inverted U-shape with intrusiveness and a U-shape with overdependence, and that users cope differently with each — distancing themselves from intrusive systems while maintaining closeness with those they over-rely on.

Blogs:

• AI’s Economic Winners argues that leadership in frontier models won’t automatically bring the biggest economic gains, leaving room for chokepoint economies like Taiwan and the Netherlands and fast-followers like Europe and India.

• The Alan Turing Institute reports on a sprint in which a small open-weight model, fine-tuned with retrieval and reasoning techniques, reached near-frontier performance on health queries while running on a laptop.

• Dr Stefaan Verhulst draws lessons from corporate AI-readiness guidance for open data stewards, chiefly governing data at the fragment rather than the dataset level, while warning that public-interest work must serve equity, not just cost.

• ODI’s Bright Simons argues that today’s AI automates verbal reasoning like law and admin while the skills behind manufacturing go under-served, risking a “premature deindustrialisation” for Africa.

• Eryk Salvaggio from Tech Policy Press contends that the UN Scientific Panel defines AI as a “system from nowhere” that erases the developers and design choices behind it, carrying a blind spot on human accountability into policy.

• OECD.AI explains how quantum computing, sensing, and communication could eventually support AI, while noting that quantum-inspired techniques are already delivering practical efficiency gains on conventional hardware.

• MIT Technology Review reports on research finding that LLMs formed their own stereotypes in a simulated hiring game, segregating candidates by group more than humans did, with reasoning models showing stronger biases.

• UKTN reports that new Prime Minister Andy Burnham has scrapped DSIT, folding its work into a renamed Business Department, despite industry warnings the move could slow the AI Opportunities Action Plan.

• Rina Chandran – Rest of World editor argues that government equity stakes in AI companies risk weakening oversight and the ability to regulate, suggesting startup funds or AI safety institutes as better ways to serve the public.

• Rest of World reports on a new Chinese marketplace where people license their likeness to AI producers, which proponents say gives people control over their image but experts warn cannot eliminate the long-term risks of biometric exploitation.

• Rest of World reports on communities from Pakistan to Africa forming data collectives to pool datasets, particularly in low-resource languages, and negotiate terms of use rather than have their data scraped without consent.

• Mohamed Shareef draws lessons from OpenAI’s Hugging Face breach, arguing that governments buying AI agents should press vendors on where an agent runs, what it can reach, who can stop it, and whether they could investigate an incident without the vendor.

• Karl Hopkins argues that the next serious public-sector AI failure may be a plausible AI-drafted summary that subtly reshapes the official record, and calls for treating high-impact drafting as record-making.

• UN Secretary-General António Guterres, opening the first UN Global Dialogue on AI Governance in Geneva, called for worldwide AI controls “worthy of global trust,” an AI Child Safety Pledge, and greater access for developing countries.

• The UK Government Digital Service sets out plans to evolve the decade-old Service Standard from a one-off assessment into a “living” framework for continuous improvement across complex, end-to-end public services.

• The State infrastructure in a box: can AI help countries import good government? asks whether AI can help developing states shortcut the slow work of building administrative capacity, while cautioning that capability is not legitimacy and imported systems raise sovereignty risks.

• A World Bank blog argues that cloud services are becoming a standardised, tradable commodity, giving countries such as Brazil, Chile, and Kenya a window to move from consumers to providers of digital services.

• Writing for the IAPP, Sarah Zhao analyses three new Chinese regulatory developments on AI ethics, AI agents and anthropomorphic AI, signalling a shift from broad principles toward more operational, risk-based rules.

• Call for Extended Abstracts for PAIRSx Public Sector 2026

• Open call for proposals, Digital Europe Programme

• Pro-Worker AI Adoption Prize, Department for Science, Innovation and Technology

• Call for Experts to join the RAISE High-Level Academic Advisory Board, The European Commission

• Innovation Fellowships 2026-27 – Route B: Policy-led, The British Academy

• AI-BOOST, an EU-funded open challenge prize programme, The European Commission

• AI Upskilling Challenge Fund, UK Department for Science, Innovation and Technology

• Chief Scientific Officer,The Alan Turing Institute

• APEX Awards, The British Academy

• The open call for proposals for New Commons Incubator for Indigenous Languages and Cultures