The Risk the Board Must Accept: Adding Agents Increases the Risk of Catastrophic Failure
Summary
A board-facing thesis, pragmatic and harsh rather than alarmist: adding agents to a normal organisation today increases catastrophic-failure risk, because reliable containment takes a security maturity and harness well above what a normal team can field. Damage tracks not only the assets handed over but the privileges of the host the agent runs inside — and the irony is that agents only become useful once given exactly that access, which is why agent projects quietly die between demo and production. The move: get the risk accepted first — on the register, owned, time-bound — and only then design the mitigation.
Key concepts
- Host-privilege blast radius — a benign task on an over-privileged host inherits that host's blast radius
- The effectiveness-requires-access irony — the access that makes an agent useful is the access that makes it dangerous
- Accept before mitigate — an unaccepted risk cannot be managed; acceptance is the precondition, not the conclusion
Key ideas
- This is not fear, uncertainty and doubt — the early-warning incidents are named and already landing.
- Even a benign agent gone over-enthusiastic can burn money, destroy data, leak data, and make decisions the business would never underwrite.
- The gap between a good demo and an un-underwritable production impact is where agent projects die.
On this site
Curated under 'what happens when one is compromised?' in the collection; the host-privilege framing feeds the Hope section's closure.