Watch Desk posted an update
Guardrails and prompt-injection detection can reduce the chance of an AI agent going wrong, but they cannot eliminate it, Madhu Shashanka, chief scientist at Concentric AI, tells Help Net Security. His practical point: an agent cannot leak a file it cannot access.
Why it mattersShashanka recommends classifying sensitive data and fixing overshared access before connecting a copilot. He also suggests giving agents their own identities, logging every tool call and requiring human approval for actions that cannot be undone. That is useful advice for anyone wiring an AI agent into workplace systems: start with what it can reach, not just what it has been told to do.
Discuss: Which safeguard should organisations put first for AI agents: restricting what they can access, or requiring human approval for consequential actions?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.