Section
Governance & Scholars
Government and regulator governance, compliance and guides, and academic research from scholars: regulation, ethics opinions and research, with the standing guides pinned first.
14 stories
Most obligations under the EU AI Act became applicable on 2 August 2026, two years after the regulation entered into force. The date covers Annex III high-risk systems, transparency duties and the national penalty regimes member states were required to lay down. Obligations for high-risk systems embedded in products regulated under Annex I follow on 2 August 2027.
Read the long piece on LexRegisterAn academic team published a benchmark measuring citation accuracy in AI legal research tools, testing seven commercial products against a set of two thousand questions drawn from state and federal practice. The paper reports error rates varying by an order of magnitude between products and finds that retrieval-augmented systems reduce but do not eliminate fabricated citations. The authors released the question set and scoring code, and invited vendors to submit corrections before a second round of testing later this year.
Read the long piece on LexRegisterCloud switching obligations under the EU Data Act became enforceable, requiring providers to remove contractual, commercial and technical barriers to moving a customer's data to a competing service. Egress charges must be withdrawn entirely from January, and providers must publish the formats in which data is exported. Legal technology vendors hosting client matter data in the Union are in scope regardless of where the vendor is established, and several vendors have already republished their standard terms to reflect the change.
Read the long piece on LexRegisterColorado lawmakers amended the state's Artificial Intelligence Act in a special session, narrowing the definition of consequential decisions and delaying the duty to conduct impact assessments by twelve months. Developers of general-purpose systems are carved out unless they market a system for a consequential use. The attorney general retains exclusive enforcement. The amendments take effect immediately on signature.
The European Commission published the final general-purpose AI code of practice, setting out how providers of large models can demonstrate compliance with the AI Act's transparency and systemic-risk obligations. Signatories gain a presumption of conformity; non-signatories must show equivalent measures. The code covers training-data summaries, copyright policy and incident reporting. Fifteen providers have signed. The Commission said the code is voluntary but that supervision begins immediately.
Read the long piece on LexRegisterThe Florida Bar issued an opinion on supervising generative AI, treating an AI tool as a nonlawyer assistant for the purposes of the supervision rule. Lawyers must verify AI output before filing or advising, and a firm must have written policies covering permitted tools and confidentiality. The opinion does not require client consent for internal use but requires it before client data enters a third-party tool.
Read the long piece on LexRegisterThe State Bar of California updated its practical guidance on generative AI, adding sections on billing, supervision of contract lawyers using AI, and disclosure to courts. The update states that a lawyer may not charge for time not spent, and that efficiency gains belong to the client where fees are hourly. The guidance remains advisory rather than a rule.
Ofcom confirmed that law firm websites offering user-to-user messaging fall within the Online Safety Act's scope where the messaging is available to the public. Firms operating client portals behind authentication are outside it. The regulator published a decision tree for borderline cases and said enforcement will prioritise services with the largest reach rather than professional services sites.
A New York State Bar Association committee addressed whether entering client information into a generative AI service is a disclosure under the confidentiality rule. The opinion says it is, unless the provider is contractually barred from training on the input and from human review. Standard consumer terms do not meet that bar. Firms are advised to obtain enterprise agreements before use.
An Ohio advisory opinion considered whether a consumer-facing legal AI product constitutes the unauthorized practice of law. The opinion finds that supplying general legal information is not, but that applying law to a user's specific facts and recommending a course of action is, regardless of disclaimers. It notes that enforcement remains a matter for the state supreme court.
Utah's Division of Consumer Protection issued guidance on when a regulated occupation must disclose that a consumer is interacting with generative AI. The guidance treats a prominent, plain-language notice at the start of an interaction as sufficient, and says a disclosure buried in terms of service is not. Legal services are named among the regulated occupations covered. The division said a first violation will be treated as an opportunity to cure, provided the interaction caused no demonstrable consumer loss.
An Illinois State Bar Association advisory opinion concluded that a lawyer may pass through the cost of a generative AI subscription as an expense only where the client agreed in advance and the charge reflects actual cost. Marking up the subscription is prohibited. The opinion draws on the existing rule governing charges for photocopying and research databases.
NIST released a draft profile applying the AI Risk Management Framework to legal services, mapping the framework's govern, map, measure and manage functions onto tasks such as document review and legal research. The draft is open for comment for sixty days. It is guidance rather than a rule, and carries no independent enforcement mechanism.
A study of bar examination performance found that candidates permitted to use a generative AI assistant scored higher on multistate performance test items but not on multiple-choice questions. The authors caution that the sample was small and drawn from a single jurisdiction's practice examination. Examiners have not indicated any change to permitted materials.