cw.is / independent notes + experiments
A working notebook about data, AI, digital health, software and the occasional rabbit hole.
/ latest transmission
Views my own // Linkedin
AI security is not simply a password and a vendor contract. Public organisations need controls for untrusted content, data exposure, connected tools and incident response.
Public-sector AI needs to demonstrate real outcomes—not merely faster outputs. A strong assurance approach tracks quality, fairness, cost, staff experience and sustainability.
Human-in-the-loop governance fails when staff cannot question, override or explain AI advice. Here is what meaningful oversight looks like in a public service.
Before putting case files, correspondence or service-user information into an AI tool, public bodies need answers on lawful purpose, minimisation, vendor access and retention.
I’ve been experimenting with a small web game recently, although calling it a game is perhaps stretching the […]
The Algorithmic Transparency Recording Standard is more than a publication form. Used well, it helps people understand why an algorithm is there and how it affects decisions.
Financial markets rarely get into serious trouble because of one isolated event. More often, problems emerge when several […]
ISO/IEC 42001 can give public bodies structure, ownership and review routines for AI—but certification does not make a particular model automatically lawful or safe.
AI use across NHS organisations is likely to be more widespread than many leaders realise. Staff are already […]
/ index fragments