Search
Adversarial Review: The Six Lenses That Halted a Rollout
Six lenses, live state, eighteen risks. I thought the safety work was shipped. The review proved three assumptions were theater. The durable rule: the model proposes; the system owns identity, trust, and state.
“We shipped the safety work” is not a status. It is a claim that needs a line check against live systems.
Before I handed a shared, governed knowledge system to a team, I ran a six-lens adversarial review against the actual box, not the design doc. Three assumptions that looked closed failed under that pressure. Eighteen risks got gated. Every durable fix that followed enforced one boundary: the model proposes; deterministic code owns identity, trust, and durable state.
If you are about to roll a multi-user AI system to people who will trust it, this is the review I wish I had run one sprint earlier.