Sort Evidence by What It Proves
Classify red-team evidence by whether it supports reproduction, impact, scope, or mitigation.
Sort each evidence item by the claim it supports most directly. Reproduction evidence Impact evidence Scope evidence Control evidence Exact user prompt, model version, retrieval IDs, and timestamp for the failing run. Policy owner confirms the generated refund threshold would violate the customer contract. Failure appears in 7 of 20 runs when the injected PDF is included, and 0 of 20 without it. After tool gating, the same injected fixture no longer changes recipients across 50 runs. A raw screenshot of the unsafe answer captured before the output was deleted. Customer-success lead estimates the draft could trigger a contractual escalation…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in