FairByDesign Reporting
agentic AI
All reportingInvestigationsMercury BriefField papersGovernance resourcesNews
-
The Preference Poisoning Problem
Part 4 of The Evidence Chain. Thirty-one companies were caught planting hidden prompts that tell AI assistants whom to trust. Your agent stays loyal, just to a version of you…
-
The Representation Gap
A US appeals court ruled that when an AI agent shops on Amazon, the user is the one acting. That protects your choice of tools, and puts every choice the…
-
The Authorization Gap
She asked her AI agent to suggest what to delete. It started deleting. When you click “allow,” can anyone prove the action that ran was the one you approved, using…
-
The Reconstruction Gap
When an AI evaluation agent breached Hugging Face in July 2026, the victim reconstructed it before the lab did. Who holds the record of what an AI did, and why…
-
MCP Enforces at the Boundary. It Cannot Vouch for Itself.
which governance functions can the protocol actually hold, and at exactly what point do you stop trusting it?
-
MCP Is an Operational Surface. Governance Has to Catch Up.
FairByDesign — Operational AI Governance Series Field Signal Model Context Protocol, or MCP, is mostly discussed as an integration breakthrough. That is accurate, but incomplete. Anthropic introduced MCP on November…
-
Classify Against the 19 May Draft Guidelines Now — Before They Are Final and Before You Are Rushed
The Commission’s draft high-risk guidelines are the clearest answer yet to “is my system high-risk?” The consultation closes 23 June. Here is how to run the test this week.