AI Regulation Weekly Brief – Week Ending 25 July 2026

This week's regulatory moves span age limits for AI companions, a last-minute reset of the EU AI Act timeline, a landmark US state law being replaced before taking effect, a major publishers' lawsuit against Google, India's zero-tolerance stance on AI-hallucinated precedents, and the Gulf's rapid build-out of judicial AI infrastructure. Together, they highlight an emerging pattern: fast convergence on child protection and judicial integrity, and increasing divergence on how far AI can be embedded in public services.

1. UK Plans Age Limits For AI Companion Chatbots

The UK Government has signalled that romantic companion chatbots will face a minimum age of 18, with similar intimate functionality on AI chatbots more broadly restricted for users under 18. Enforcement will sit with Ofcom, with detailed regulations expected around spring 2027, placing the UK in a growing coalition of jurisdictions focusing on protecting minors from AI companionship.

For AI product teams, this shifts youth-risk management from a nice to have to a regulatory obligation, especially for any AI system that simulates intimacy or emotional dependency. Compliance planning now needs to incorporate robust age-assurance, functionality gating, and clear design standards for under-18 experiences.

2. EU Quietly Resets The AI Act Clock

The EU's Digital Omnibus on AI, signed on 8 July and awaiting publication in the Official Journal, effectively rewrites the AI Act's implementation timeline just days before core provisions start to bite. High-risk obligations for stand-alone Annex III systems are pushed to 2 December 2027, and product-embedded Annex I systems to 2 August 2028, while a new ban on AI-generated non-consensual intimate imagery is added to Article 5.

Crucially, transparency duties under Article 50, covering chatbot disclosure, synthetic-content marking and deepfake labelling, along with the Commission's powers over general-purpose AI models, still take effect on 2 August 2026. AI providers now face two clocks: immediate pressure on transparency and model governance in 2026, with heavier high-risk system obligations following in 2027-28.

3. Colorado Rewrites Its Landmark AI Law Before Day One

Colorado's pioneering AI Act, originally due to take effect on 30 June 2026, was repealed before it ever applied and replaced by SB 26-189. The new law drops the broad algorithmic discrimination regime in favour of three narrower duties for automated decision-making deployers: pre-use notice, adverse-outcome processes including human review and correction rights, and three-year record-retention obligations enforced by the state attorney general from 1 January 2027.

The episode is a cautionary tale for compliance teams building programmes around untested statutes. It also feeds into Washington's debate over federal pre-emption of state AI laws, signalling that US state experimentation will remain fluid and politically contested.

4. Publishers Take The AI Training Battle To Google

A group of major publishers, including Hachette Book Group, Cengage Learning, Elsevier and author Scott Turow, have filed a proposed class action against Google in the Southern District of New York. The suit alleges that Gemini was trained on books and journal articles supplied to Google Books, Play and Scholar for limited purposes, as well as scraped works, asserting direct and contributory infringement and removal of copyright-management information.

What stands out is the legal theory: after key fair-use wins for AI training in 2025, the argument has shifted from training is theft to training exceeded the licence. This contract-adjacent framing is harder to neutralise with fair-use defences and moves the dispute closer to how publishers actually license their databases, with clear implications for any AI developer relying on licensed corpora.

5. India's Supreme Court Draws A Hard Line On AI-Hallucinated Precedents

India's Supreme Court has set aside tribunal orders after discovering they relied on non-existent, AI-generated precedents, holding that citing unverified AI output constitutes misconduct for advocates and a serious lapse for adjudicators. The Court has directed the Bar Council of India to frame norms with disciplinary consequences, as its draft Regulations for Use of AI in Courts move toward finalisation following a mid-July consultation deadline.

The draft rules would confine AI to assistive roles, explicitly barring algorithmic decision-making in judicial outcomes and prohibiting AI risk-scoring of bail eligibility or witness credibility. India is thus sketching a clear settlement: AI can support administration and research, but not substitute human reasoning in adjudication, with verification of AI-assisted material as a non-negotiable duty.

6. Abu Dhabi Accelerates Judicial AI Deployment

The Abu Dhabi Judicial Department's AI committee has reviewed a slate of projects, including a legal judgments assistant, a smart assistant for notaries, AI-based complaint analysis and routing, prosecution system upgrades and court performance dashboards. Rather than solely regulating AI, the Gulf's judicial systems are actively embedding it into court infrastructure, pushing a procedural-modernisation track that is ahead of many larger jurisdictions.

This dual reality, governance in some courts and deployment in others, underscores that AI and the courts now covers both how judicial systems control AI and how they use it themselves. For vendors and policymakers, the Gulf provides a live test bed for what responsible judicial automation might look like at scale.

Why These Moves Matter For Global AI Governance

Taken together, this week's actions show rapid convergence on three themes: protecting minors from AI-mediated intimacy, demanding verifiable inputs to judicial decisions, and clarifying how far state-level AI laws can reach before being reset or pre-empted. At the same time, they expose real divergence on institutional appetite for deploying AI in core public functions, from courts to regulators.

For builders, banks, and policy teams, the practical implication is clear: global AI compliance will track different timelines and risk appetites, but certain lines, children's safety, judicial integrity, transparency around synthetic content, are solidifying fast. Monitoring these shifts weekly is no longer optional for anyone operating AI systems at scale.

About The World AI Regulation Summit

This Weekly Brief is published by the World AI Regulation Summit, a global forum dedicated to sharing best practices across jurisdictions on AI governance, compliance and policy design. The annual meeting will be held in London on 5-6 November 2026, convening regulators, industry leaders, civil-society experts and technologists.