What’s happening at K-AI
Events, partnerships, talks — K-AI in the French and European AI ecosystem.
A document review campaign can close at 100% without reviewing what your AI reads
A review campaign closed at 100% covers its own scope, not what your assistant reads. A "read by AI" column puts the two lists side by side.
When AI fills in the status column, the agent reading it may see a decision nobody made
An AI-filled status column is a reading of the document. The agent that reads it sees a decision. A provenance column says who validated it.
Dead evidence never changes, and the control that cites it can stay green
A control reassessed when its evidence changes stays green if the procedure never moves while the organisation it describes has changed.
The knowledge layer you are budgeting for 2027 will inherit the state of your procedures
A knowledge layer reads your procedures as they stand. Without an entry baseline, their defects become its own, with no record of where they began.
A withdrawn document can keep answering through your AI
A withdrawal procedure handles the file, not the copies AI made of it. The withdrawal record checks that a withdrawn document is really gone.
When your AI agent contradicts itself, stabilising it can hide two documents that contradict each other
Two answers to one question: the cause may be the model, retrieval or two contradictory documents. The double-answer log tells them apart.
Experts fix their documents when the AI gets an answer wrong citing them. Only if they find out
A wrong-answer report usually stops at the AI team. The source referral carries it back to the expert who can fix the cited document.
Your AI agents now have an owner. The documents they cite still don't
Glean enforces a single owner per agent, Gartner makes ownership a requirement. The documents those agents cite still have no named accountable owner.
The description of your documents already exists somewhere in your company. Your AI programme is rebuilding it
Gartner: through 2028, building your own unstructured metadata will cost 300% more than reusing what exists. The trade-off is settled at framing.
Your tools catch contradictions inside a package. Nobody counts the ones that cross packages
Two buyer-side corporate funds just financed automated document review. At package scale. The cross-functional document falls outside every perimeter.
An incorrect answer to a regulator is a separate infringement from the one it is asking about
Article 99(5) fines an incorrect reply: €7.5M or 1% of turnover. And that reply is assembled from documents that contradict each other.
Authoritative sources for enterprise AI: what you certify when you designate one
SharePoint now lets you mark a site authoritative for Copilot. The signal sits on the container; the contradiction lives in the documents.