The Week in AI.
AI got cheaper this week. Governing it did not.
The Chartered Insurance Institute warned this week that AI governance across insurance is running ahead of staff capability, and that a human in the loop is not a safeguard unless that human is active, informed and accountable. In the same week Google halved its model prices again. For claims and legal teams the constraint is no longer licence cost. It is whether the person who signs the output knows enough to disagree with it.
Read the briefingThree AI labs breached real companies. The test conditions are the story.
Meta became the third major lab, after Anthropic and OpenAI, to disclose that one of its models breached a third party during testing. In all three cases the same independent testing firm gave the models internet access in error. In the same week the UK's own institute published rogue-agent findings from runs where safeguards had been deliberately disabled. Neither story means what the headline says on its own.
Read the briefingYour firm's working methods are now files that travel. Almost nobody is checking where they came from.
The written procedure, not the model, is where a claims or legal operation's real know-how now sits. Those procedures have become a portable file format that more than 30 tools can read, share and run. The audits say a large share of the ones already circulating are not safe to run.
Read the briefingAI stopped waiting for instructions this week and started logging in. Access is now the control question.
The tools got hands. Claude can now open a browser and act across live systems, and a new integration lets it log in without ever seeing the password. For a claims or legal operation, that shifts the risk from what an assistant writes to what it can reach. The question this week is not is the draft any good, it is who gave the tool the keys, and can you show what it did with them.
Read the briefingAI reached deeper into the file this week. Custody of that file is now a live question.
Every major release this week moved AI closer to the systems where case work actually lives. Files, inboxes, calendars, task boards. For most industries that is a convenience story. For claims and legal teams it is a custody story, and custody is a duty you have to be able to evidence.
Read the briefingAI split into tiers this week. The job is matching the model to the matter.
Anthropic shipped a workhorse and a flagship in the same week, and the advice from every quarter was to plan with the best model and execute with cheaper ones. For claims and legal operations, where volume correspondence sits alongside high-stakes judgement work, tiering is not a technical detail. It's the cost base.
Read the briefingNew edition every Wednesday. Specialist workflows. Compliance first.