01
The CII says a human in the loop is not a safeguard on its own.
The Chartered Insurance Institute warned of a critical AI fluency gap across insurance and personal finance, following a roundtable it convened in June with Chartered firms, technology providers and academia. Its report, Responsible AI: from policy to practice, argues that boards, risk functions and customer-facing professionals need training well beyond basic tool use, building professional scepticism and the confidence to question automated outputs. It warned against relying on a human in the loop as a safeguard in itself.
CaseFlow angle. Name the person who signs off each AI-assisted output on your files. Then ask them to talk you through one they approved last week, in order, and say what they checked. If they cannot, you do not have a control, you have a signature. Record the answer, the date, and the training gap it exposes.
02
Model prices halved again, three weeks after the last release.
Google shipped Gemini 3.7 Flash on 14 August at 75 cents per million input tokens and 3.75 dollars per million output tokens, half the previous rate, held at that level until the end of December. Grok 4.6 arrived with more capability at unchanged pricing. The cheap, fast tier is where most claims and legal document work actually runs.
CaseFlow angle. Take the document job you costed last year and rejected as too expensive to run at volume. Re-price it at this week's rates before the introductory pricing ends in December, and put the new number in front of whoever holds the budget.
03
Claude now works inside the browser, so a missing integration stops being the reason.
On 12 August Claude's Chrome side panel became a full Cowork session. It can see the page, click, type, navigate and fill forms using existing logins, and a task started in a tab can be finished on desktop or mobile. Anthropic names internal dashboards, legacy systems and vendor portals as the target. Live on Max and Team, rolling out to Pro.
CaseFlow angle. List the three insurer or court portals your handlers retype into. For each one, write down whose login would be used, what a wrong submission would cost, and who signs before it goes. An agent acting on a handler's login leaves the handler's audit trail, not its own, so settle that before you automate a single field.
04
A video model now treats a PDF as raw material.
Alibaba's Wan 3.0 entered public beta on 6 August, generating clips of up to 30 seconds from text, images, audio, web pages, PDFs and slide decks, with up to 20 assets in a single call. Grok's Imagine Image 2.0 pushed still-image quality up over the same period. Convincing moving footage no longer needs a filming step.
CaseFlow angle. Write down this week how your team tests a photo or video that arrives as evidence. Who checks where it came from, what metadata you keep, and at what point you ask for the original device file. If the honest answer is that nobody checks, that is the gap, and writing the step down costs nothing.
05
The FCA's latest value measures data has motor paying out a bigger share of a smaller pot.
The FCA's general insurance value measures data for 2025 puts motor's claims ratio at 59.1 percent, up from 54.2 percent the year before, the sharpest year-on-year move of any product line it tracks. Average motor premiums fell about 7 percent over the same period. A 59 percent ratio leaves far less headroom to absorb repair inflation, theft or regulatory change.
CaseFlow angle. This is the number sitting behind every hire-rate and repair-cost challenge you will receive this year. Thinner headroom means sharper scrutiny of duration and need, not softer. Make sure the hire period on every open file is evidenced on the file as it runs, rather than reconstructed at the point of challenge.