Definition
Plain language
Stacking several independent safeguards so that if one fails, the next one still catches the problem.
As stated in the literature
A security design principle of layering multiple independent protective mechanisms so no single failure compromises the whole system; applied to AI-agent safety by guarding actions at several lifecycle stages, and to verifier stacks like MaxProof's four-layer grader.
Also called: defense in depth
Why it matters: It matters because relying on a single safeguard means one slip-up lets everything through, while layers keep protecting you when one fails.
For example, an AI agent might be checked before it plans an action, again before it runs it, and once more before results are returned.
Heard on the show
“… The new verifier they built for this generation is four layers of defense-in-depth, and what I love about it is the one-to-one mapping: every single layer exists because of a …”Episode 133 — How MiniMax Turned a Reward-Hacking Disaster Into Olympiad Gold