After the first fines for fake AI citations, the CGAE is already drafting a confidentiality circular

Receiving client enquiries accompanied by documents generated with artificial intelligence "is already a standard". The statement comes from Manuel Mata, president of the Ethics Committee of the General Council of Spanish Lawyers (CGAE) and dean of the Castellón bar, in an interview published today by El Español that takes stock of where things stand: sanctions already imposed, an ethics framework under construction, and confidentiality as the next front.

The sanctions that set the starting point

The tally is familiar to this outlet's readers. In February 2026, the High Court of Justice of the Canary Islands fined a lawyer 420 euros for including 48 non-existent or incorrect AI-generated references in an appeal; in July, the High Court of Justice of Galicia imposed 1,800 euros on another lawyer for 24 invented citations and texts. Both cases add to the warning the Constitutional Court issued in 2024 to a lawyer for citing entirely false rulings of that very body. We covered these precedents in our analysis of the fines for AI hallucinations in the Canaries and Galicia and of the fake case law and sanctions framework.

The framework under construction

In January the CGAE presented its White Paper on Artificial Intelligence and the Legal Profession, which noted a striking gap: 60% of legal services professionals already use AI applications, but only 8% say they know how they work. In April it approved Interpretative Circular 3/2026 of the Code of Ethics, focused on drafting, signing and filing documents with generative AI: diligence, professional integrity and human oversight, with AI as an auxiliary tool and never a substitute.

The new element in the interview is the roadmap: the circular expressly left out confidentiality and professional secrecy, which will be the subject of their own circular. Mata sums up the criterion that will guide it: "The duty of diligence does not end with reviewing the generated content; it also requires knowing what happens to the information entered into a tool". He also warns of the limits of anonymisation: "Hiding names does not always remove all identifying information, because data can remain in the metadata or in other parts of the document".

On the other side of the bench

AI use is not foreign to judges either. The General Council of the Judiciary approved Instruction 2/2026, which sets limits on its use by judges and magistrates, and in April it fined a magistrate 1,000 euros for disclosing information by entering a court file into an AI platform. At European level, the AI Act classifies as high-risk certain systems intended to help judicial authorities research and interpret facts and legal rules.

The president of the Spanish Bar, Salvador González Martín, calls in the same report for an exercise in governance: "Technology is advancing at a speed that imposes rhythms that do not allow us to reflect". Mata translates it into professional terms: the lawyer "is responsible to their client for that document being right, for it being properly reviewed, regardless of the tools used to prepare it".

What firms can do while the circular arrives

Without waiting for the final text, the criteria already sketched allow action: inventory which tools are used and under what data-processing terms, define what information may be entered into each one, systematically verify citations and statutes before signing, and record the workflow followed on each matter. Our guide to contract review with AI and human control and the one on legal prompting for law firms develop those controls in daily practice.