Welcome to the first edition of our monthly legal-AI roundup: the regulatory, judicial, and market developments that mattered, in one page, with sources you can open. July 2026 was dominated by calendars - the EU rewrote its AI Act timeline weeks before the deadline everyone had planned around, and India's Supreme Court turned citation hygiene into consequence. Everything below is as of 19 July 2026; pending items may move.
EU: the Digital Omnibus resets the AI Act calendar
The Digital Omnibus on AI - the amendment package renegotiating the AI Act's timeline - was signed on 8 July 2026 and now awaits publication in the Official Journal. It defers the high-risk regime by more than a year: stand-alone Annex III systems move to 2 December 2027 and Annex I product-embedded AI to 2 August 2028, per Travers Smith's briefing. What did not move: the Article 50 transparency duties - chatbot disclosure, synthetic-content marking, deepfake labelling - and the Commission's enforcement powers over general-purpose AI models, both of which take effect on 2 August 2026.
| Obligation | Date |
|---|---|
| Article 50 transparency duties (chatbot disclosure, content marking, deepfake labels) | 2 August 2026 |
| Commission enforcement powers over GPAI models | 2 August 2026 |
| Machine-readable content marking for systems already on the market | 2 December 2026 |
| High-risk obligations - stand-alone Annex III systems | 2 December 2027 |
| High-risk obligations - Annex I product-embedded systems | 2 August 2028 |
Why it matters: AI Act advice now runs on two clocks - transparency in weeks, high-risk in 2027-28 - and the underlying obligations are unchanged: deferral is breathing room, not relief. Our wave-by-wave guide to what became enforceable on 2 August 2026 sets out the corrected timeline, and our explainer on the EU AI Act for law firms covers the deployer duties that survive the reshuffle.
US: Colorado replaces its landmark AI law before it ever applied
30 June 2026 - the day Colorado's pioneering AI Act (SB 24-205) was due to take effect - passed without it ever applying. Governor Polis had signed SB 26-189 on 14 May, repealing and replacing the 2024 law with a narrower framework effective 1 January 2027, as analysed by Littler and Finnegan. The replacement swaps the broad algorithmic-discrimination regime for three deployer duties around automated decision-making technology: pre-use notice, an adverse-outcome process with correction rights and meaningful human review, and three-year record retention, enforced by the state attorney general.
Why it matters: the first comprehensive US state AI law was rewritten before its first day in force - the state patchwork is still fluid, and compliance programmes built on draft-era assumptions need quarterly re-checking.
India: zero tolerance for fake precedents, and draft rules for AI in courts
On 2 July 2026, a Supreme Court bench of Justices P.S. Narasimha and Alok Aradhe set aside NCLT and NCLAT orders after finding they relied on non-existent, AI-generated precedents, calling hallucinated authority "catastrophic" to the judicial process and demanding a zero-tolerance approach to citing AI output without verification. The Court held that citing unverified AI-generated judgments is misconduct for an advocate and a serious lapse for an adjudicator, and directed the Bar Council of India to constitute a committee and prescribe norms with disciplinary consequences.
Why it matters: India's highest court has converted citation hygiene from guidance into misconduct - and applied the same standard to tribunals that rely on such material as to the lawyers who file it.
The ruling landed mid-consultation on the Supreme Court AI Committee's draft Regulations for Use of Artificial Intelligence in Courts, 2026, published in June, with comments closing 15 July. The draft confines AI in courts to an assistive role: no judicial outcome may rest on algorithmic decision-making alone, and AI may not be used for risk scoring - bail eligibility, flight risk, recidivism, or witness credibility - with opaque systems barred from any process affecting rights or liberty.
Why it matters: together, the judgment and the draft regulations sketch India's emerging settlement - administrative and research uses in, adjudicative uses out, verification always - a framework we track in our India mid-year update.
UK: AI in expert evidence moves up the rules agenda
The debate over rules for generative AI in English court documents advanced, with Norton Rose Fulbright analysing in July the responses to the Civil Justice Council's consultation on AI in preparing court documents. The Bar Council and the Law Society responded in April and May; a leading proposal would amend the expert's statement of truth to require disclosure of what AI was used and how, with the Law Society pressing for red-line court guidance and arguing that even transcription and translation carry disclosable risk.
Why it matters: expert evidence is where undisclosed AI use meets independence duties head-on, and a disclosure requirement in the statement of truth would make AI use in litigation formally auditable for the first time.
MENA: Gulf courts and firms build out legal AI
The Abu Dhabi Judicial Department's Artificial Intelligence in Judicial and Notary Services Committee, chaired by Undersecretary Counsellor Yousef Saeed Al Abri, reviewed a slate of AI projects in early July: a Legal Judgments Assistant, a smart assistant for notaries, AI-based complaint analysis and routing, Public Prosecution system upgrades, and court performance dashboards, per the department's statement.
Why it matters: the Gulf's judicial systems are adopting AI as court infrastructure - service efficiency, routing, notarial work - a procedural-modernisation track that runs well ahead of many larger jurisdictions.
On the private side, Dubai firm Horizons & Co and RIT Dubai publicised their partnership to develop what they describe as the first legal AI built in the Middle East - announced 1 June and detailed on 17 July - a tool designed around UAE law rather than adapted from foreign-trained systems.
Why it matters: jurisdiction-native training is becoming a selling point in markets whose civil-law and free-zone structures are poorly served by tools built elsewhere - a theme from our Gulf and Singapore adoption snapshot.
Funding: legal AI mints another unicorn
Norm Ai raised a $120 million Series C at a $1.2 billion valuation on 7 July, led by Khosla Ventures, taking its total raised past $260 million roughly three years after founding. The company builds compliance-focused AI agents, supervisory agents that oversee other agents, and Norm Law, an AI-native firm staffed by supervising attorneys. LawSites notes it joins the legal AI unicorn tier alongside Harvey - $11 billion after its March Series G - and Legora at $5.6 billion.
Why it matters: capital is concentrating in vendors that pair AI agents with human supervision structures - the model regulators keep signalling they can live with - and the unicorn tier is now several companies deep.
Products: agentic legal AI reaches general availability
Intapp made Celeste generally available in mid-July - an agentic "AI coworker" for the business side of law firms, covering deal screening, conflicts clearance, business development, and lateral hiring, with a playbook library and a no-code builder, after an early-adopter release at firms including BakerHostetler, per LawSites.
Why it matters: the agentic wave is reaching the operational core of firms - intake, conflicts, laterals - where errors have regulatory consequences, which will test whether agent governance is ready for production rather than pilots.
Litera, meanwhile, relaunched the company around Lito, its AI agent, expanding it across existing workflows ahead of general availability at ILTACON in August, alongside a wider rollout of its Foundation 365 knowledge platform.
Why it matters: when an installed-base vendor of Litera's size reorganises around one agent, agentic drafting becomes the default roadmap for the tools firms already own - a pattern from our shipped-vs-hype audit.
Copyright: publishers take the AI training fight to Google
Hachette Book Group, Cengage Learning, Elsevier, author Scott Turow, and his company S.C.R.I.B.E. filed a proposed class action against Google in the Southern District of New York on 10 July, alleging Gemini was trained on books and journal articles supplied to Google Books, Play, and Scholar for limited purposes, plus scraped works, with claims spanning direct and contributory infringement and removal of copyright-management information, per TechCrunch. The suit follows 2025's California fair-use rulings - and Anthropic's $1.5 billion settlement over pirated source libraries - but puts the question to a New York court on a record built on long-standing licensing relationships.
Why it matters: the theory has shifted from "training is theft" to "training exceeded the licence" - a contract-adjacent framing that survives fair-use rulings and lands close to how legal publishers license their own databases.
Adoption: the operationalisation gap, quantified
The latest Law Department Operations Survey, discussed in a Law.com-covered webinar on 14 July, found that 23% of legal departments have fully operationalised generative AI, 33% report cost savings, and 64% report tangible non-financial benefits - while usability has overtaken security as the top AI concern, with "shadow AI" use outside approved systems a growing worry. The panel's recurring diagnosis: departments pick tools before defining the business problem, and few set baselines that would later prove savings.
Why it matters: the adoption question has moved from whether to how well - and the gap between individual use and operationalised value is now the number to manage, a measurement problem we cover in our legal AI ROI framework.
What to watch in August
Three dates anchor the next edition. 2 August: the EU AI Act's transparency duties and GPAI enforcement powers take effect - the first hard test of the narrowed deadline. Mid-August: ILTACON, where Litera's Lito is slated for general availability. And in India, the closed consultation moves the draft court-AI regulations toward finalisation, with the Bar Council's citation-norms committee working alongside. The August edition picks up all three.