Thelonious ICYMI — Edition 20 — 20 August 2026
This weekly briefing on AI legal and regulatory developments frames the key theme: the accountability gradient. Three of this week’s five developments landed on the same day, 17 August 2026, and read very differently depending on where the user sits. A regulator told solicitors they are answerable for every word their tools produce. A federal court told a litigant that a judge who allegedly outsourced an entire ruling to AI cannot be sued for it. Between those two positions sits a widening gap, and Edition 20 tracks who falls into it.
Lead Stories
- The Misuse of AI: SRA warning notice — The Solicitors Regulation Authority issued a warning notice on 17 August 2026 applying to all SRA-regulated firms and individuals. Two risks are identified. The first is hallucination: fabricated cases, false citations and fictitious references that read credibly and end up in court documents. The second is confidentiality, where client information entered into public AI tools puts confidentiality, legal professional privilege and data protection obligations at risk. The operative statement is unambiguous: “AI has no separate legal personality; solicitors and regulated individuals who use AI in the course of delivering legal services remain accountable for their work and outputs.” Use of a tool is not a defence to an inaccurate submission. The notice cites Principles 1, 2, 4, 5 and 7, alongside Code of Conduct provisions on misleading the court, competence, supervision, confidentiality and contempt. Firms are directed to verify citations before filing, maintain human oversight and professional judgement, supervise staff using these tools, check contractual and technical safeguards where client data is involved, and hold governance structures capable of managing the risk. The significance is jurisdictional rather than technical: AI governance has been placed inside professional conduct, not inside IT procurement.
- Phillips v. Parlade — Decided 17 August 2026 in the US District Court for the District of Nevada before Judge Gloria Navarro. The plaintiff alleged that a state court judge “relied wholly on artificial intelligence to issue a judicial ruling, without any discretionary human thought.” The court applied the standard test for absolute judicial immunity and found the act judicial in nature on three grounds: issuing rulings is a normal judicial function, the matter concerned a case properly pending before the judge, and nothing alleged conduct outside chambers. Immunity applied even taking the allegations as true, and the case was closed. Objections of this kind, the court held, belong in state appellate review, mandamus or judicial disciplinary proceedings, not in a federal civil action. Read alongside the SRA notice, the asymmetry is the story. A solicitor who files an AI-generated citation faces regulatory and civil exposure directly. A judge who allegedly generates the ruling itself is answerable only through slower and considerably less transparent channels.
- Round Hill Music v. Anthropic — Filed 17 August 2026 in the US District Court for the Northern District of California, alleging that Anthropic used data-scraping software to circumvent technological protection measures and reproduce copyrighted song lyrics to train its Claude models. The complaint pleads direct infringement under the Copyright Act alongside DMCA violations covering circumvention of access controls and removal of copyright management information. Five hundred compositions are pleaded as a prioritised representative bellwether, with the publisher signalling amendment to cover potentially ten thousand or more. Round Hill controls 14,364 compositions and 16,873 sound recordings. At the statutory maximum of $150,000 per wilfully infringed work, the publisher estimates exposure in the hundreds of millions and potentially at or beyond $1 billion.
- Round Hill Music v. Suno — Filed the same day in the same court, alleging improper consumption of over 500 copyrighted musical works to train Suno’s models, again pleading both Copyright Act and DMCA circumvention claims. The distinguishing feature is the co-defendant. Israeli data-scraping firm Bright Data Ltd and its US subsidiary are named on the basis that they supplied the proxy networks and scraping tools that allowed Suno to reach music held behind access controls on licensed platforms, a contributory infringement claim quantified at $75 million. Two points carry the significance. The DMCA circumvention theory sidesteps fair use by focusing on how access was obtained rather than what was done with the material afterwards. And naming the scraping supplier extends the liability map from model developers to the infrastructure layer feeding them. Round Hill has stated it intends to take both cases to trial and has instructed trial counsel rather than settlement negotiators.
- The Great Transshipment Scam — A White House report describes exporters, principally from China, routing goods through third countries to evade US tariffs, with hundreds of thousands of jobs and billions of dollars in income cited as the cost. The proposed remedy is an “AI-enabled Detective Border” capable of identifying high-risk trade flows by analysing ownership structures and shipment data at scale. The development belongs in this briefing because of what it does to AI’s position in the argument. Here it is not the regulated object but the enforcement mechanism. Trade practitioners will need to engage with questions that have until now sat at the edge of the practice: evidential standards for algorithmic determinations, the procedure for challenging one, and the remedy available when a false positive stops a shipment.