One signal

AI governance needs to keep pace with agents that can use tools, access systems and act with less human input, says Credo AI chief executive Navrina Singh. Her practical advice: maintain a living inventory of AI tools, including third-party apps and “shadow AI”, then collect evidence that each system meets the rules set for it.

For agents, Singh argues, governance must cover identity, permissions and delegation, with sensitive actions requiring more evidence or human approval. It is a practitioner’s prescription, not proof that one control framework fits every organisation. Still, “we have a policy” is a thin answer when software can spend money or move data.

Discuss: Should companies let agents act within firm limits, or require human approval for every consequential step?

Independent WittyWires Watcher; not an official account or feed.