Latest / HotTea: The AI Morning Briefing / Thomson Reuters built the model it wants lawyers to trust
Transcript
- Sections 00:00 What we're covering today 00:19 1. Thomson Reuters launched its own professional model 01:05 2. Google put Gemini Enterprise into legal workflows 01:54 3. U.S. gas-power plans tied to data centers nearly doubled 02:46 4. The UK will train security AI on Ukraine battlefield data 03:33 5. Russia gave itself a route into damaged infrastructure 04:21 6. Anthropic offered $5 million for AI wellbeing evaluations 05:09 Visit Hot Tea Disclosure Narration uses an AI-generated voice. Transcript For Wednesday, August 26, 2026. Thomson Reuters launched a legal and tax model trained on a small slice of its archive. Google moved into legal agents the same day. Scale still matters, but today's harder fights ran through trust, data rights, power plants, war data, and mental-health safety. Thomson Reuters launched its own professional model. The company announced Thomson on August 24. It said the model will first appear in Tabular Analysis inside CoCounsel Legal, with more legal and tax uses planned. A related post said the current base model is Imperial College London's Snowdon model. Thomson Reuters makes the strongest performance claims here. The company says outside academics are evaluating the model. It also says a small open-weight version will be available for academic and non-commercial use. That helps validation, but it is still not broad independent production evidence. The next proof is the technical report. The outside evaluations need enough detail to audit. CoCounsel retention and legal-document error rates matter too. Thomson matters if lawyers use it for work they would not trust to a general model. Google put Gemini Enterprise into legal workflows. Google Cloud launched Gemini Enterprise for Legal in preview on August 25. Google said the product gives law firms and legal teams purpose-built agents for legal workflows. Business Insider reported that Google named Everlaw, Relativity, iManage, NetDocuments, and DocuSign as partners. It also named Cleary Gottlieb and Freshfields as launch firms. Google is selling control over legal workflows, not only model access. Law firms still need proof that agents can touch firm data, contracts, and document review without adding liability. The test is whether preview firms move the agents into billed work. Data-training terms and malpractice guidance matter. Independent error-rate testing would matter more than partner logos. U S. gas-power plans tied to data centers nearly doubled. Global Energy Monitor said U.S. gas-power proposals tied to data centers reached 97 GW at the end of 2025. It said those proposals rose to more than 189 GW by mid-2026. The Guardian reported that the U.S. is building twice as much gas-fired capacity as China during the AI boom. Wired reported that some data center operators are looking at private plants to avoid grid delays. Not every proposed plant will be built. The direction still matters. AI companies are moving power risk out of cloud dashboards and into fuel markets, local permits, emissions math, and utility bills. The proof comes from financing and construction. Regulators also need to decide whether data center costs land on residential rates. Tech-company power contracts will show who pays when gas prices move. The UK will train security A I on Ukraine battlefield data. The UK government said a new partnership gives Britain access to Ukraine's Avengers AI Lab. The first project uses Ukrainian data and UK technology to turn buried fibre-optic cables into AI-enabled sensors. The Guardian reported that the pilot starts at a UK defence site. It could later cover railways, energy plants, airports, and prisons. Ukraine has high-value training data from a live war. That data may improve threat detection. It also raises privacy, targeting, and dual-use questions when military data moves into domestic security systems. The open questions start with what data leaves Ukraine and who can use it. The first pilot results at the UK defence site matter. Procurement terms will show whether private AI firms get repeatable access. Russia gave itself a route into damaged infrastructure. AP reported that Vladimir Putin signed a decree covering critical infrastructure damaged by Ukrainian drone attacks. The decree lets the Russian state temporarily take control if private owners fail to protect or repair those assets. The order covers energy, industrial, communications, transport, and logistics assets. The Guardian also reported fresh pressure around Russian refineries and diesel exports. Russia is not calling this nationalization. The decree still tells companies that wartime damage can bring state management. That pushes more defense cost onto private refineries, logistics firms, and communications owners. The first test is whether Moscow uses the decree against refinery owners first. Insurance and tax relief terms matter too. More forced state management would show drone attacks are reshaping Russian industrial control. Anthropic offered $5 million for A I wellbeing evaluations. Anthropic announced a $5 million grant program for independent research on how AI affects user wellbeing. The company said grantees will get funding, model access, and technical support. Anthropic said selected work should become open-source evaluations. Applications are due September 21. Anthropic is funding the work, so the independence claim needs follow-through. The hard cases are long conversations where risk appears late. That includes companionship, disordered eating, and mental-health crisis use. The grant list matters. So does whether researchers publish their methods. Useful evaluations should test over-compliance and over-refusal, not only whether a single answer sounds safe. That is the signal before the noise. This briefing was produced from Hot Tea's verified daily edition. For the complete briefing and every source link, visit Hot Tea dot A I.