01
Two new Claudes arrived: the engines behind claims and legal AI levelled up.
Anthropic released Claude Sonnet 5, a near-flagship model at everyday pricing, then Claude Fable 5, its most capable model for agents and long-running work. These are the models specialist tools in this sector, including ours, are built on. Access to the restricted tiers also returned after US export controls lifted.
CaseFlow angle. The tools your team uses are about to get noticeably better at the same job. Ask your suppliers which model tier sits under each feature you pay for, and what the upgrade means for accuracy on your file types.
02
Strong models plan, cheap models do. Sound familiar?
The week's clearest advice mirrors how a good firm already works: the senior person designs the precedent and reviews the output, the team runs the volume. Use the most capable model to write and check the process, and let cheaper models execute it. Analysts framed three lanes: cheap volume, balanced work, frontier only.
CaseFlow angle. Treat models like fee earners. Precedent letters, chasers and acknowledgements belong on the workhorse tier. Liability arguments and quantum judgement earn the flagship. Price each lane accordingly.
03
AI agents joined the task board. Supervision rules apply.
Notion 3.6 now lets teams assign tasks to Claude, Cursor and other agents on a shared board alongside human staff, with audit-log visibility for enterprise users. Agents are becoming team members you allocate work to, not tools you open.
CaseFlow angle. If an agent takes work from your board, supervise it like a trainee. Its own column, defined task types, and a named human review step before anything leaves the firm. Your regulator will expect nothing less.
04
The three powers rule: read files, ingest content, communicate. Pick two.
A widely shared governance briefing argued AI risk now comes from three powers: reading your data, taking in outside content, and communicating externally. An assistant holding all three is exposed to prompt injection and data leakage. For firms handling client files, that's a confidentiality question, not an IT one.
CaseFlow angle. Audit every AI tool that touches case data this week. If one can read files, browse the web and send messages, switch one power off. Client confidentiality is the test, convenience is not.
05
Generative AI is now forging claims evidence at record scale.
Aviva reported stopping record levels of claims fraud involving AI-generated materials, and IFED has partnered with Nominet to take down spoof insurer websites. Deepfaked images, synthetic documents and fake portals are now standard fraud tooling, and they are landing in motor and injury claims first.
CaseFlow angle. Assume some evidence on your files is machine-made. Ask what verification sits behind images and documents you rely on, and log provenance questions early. The firms that can show a verification step will settle disputes about evidence faster.