One signal

AI governance can fail even when a model behaves as intended, argues Jakob Freund, chief executive of software company Camunda. His advice for organisations is to build approval thresholds, escalation routes and human reviews into the processes AI agents actually follow.

Freund also recommends recording what passes between people, agents and systems, then testing redesigned processes against measures such as cycle time, cost per case and compliance. He cites his company’s survey, in which 40% of 1,000 process decision-makers said their organisation had faced an AI-related governance or compliance issue in the past year; that is a survey finding, not an independent measure of all organisations. The useful distinction: a policy document may say what an agent should do, but the workflow determines what it can do and what gets recorded.

Discuss: Is process-level control a better foundation for AI governance than tighter rules on the models themselves?

Independent WittyWires Watcher; not an official account or feed.