https://umbris.org/ https://umbris.org/about/ https://umbris.org/blog/ https://umbris.org/blog/authors/wishardry/ https://umbris.org/blog/authors/zourage/ https://umbris.org/blog/kydm-know-your-damn-models/ https://umbris.org/blog/v1-is-almost-here/ https://umbris.org/faq/ https://umbris.org/get-started/ https://umbris.org/learn/ https://umbris.org/learn/attacking-llm-apps/ https://umbris.org/learn/attacking-llm-apps/count-with-error-bars/ https://umbris.org/learn/attacking-llm-apps/distrust-the-scoreboard/ https://umbris.org/learn/attacking-llm-apps/one-token-stream/ https://umbris.org/learn/attacking-llm-apps/tools-are-the-attack-surface/ https://umbris.org/learn/attacking-llm-apps/untrusted-data-meets-private-data/ https://umbris.org/learn/attacking-llm-apps/write-the-finding/ https://umbris.org/learn/llm-foundations/ https://umbris.org/learn/llm-foundations/whats-a-token/ https://umbris.org/learn/red-teaming-rlhf/ https://umbris.org/learn/red-teaming-rlhf/audio-and-speech-tokens/ https://umbris.org/learn/red-teaming-rlhf/capabilities-stay-in-the-weights/ https://umbris.org/learn/red-teaming-rlhf/force-the-opening/ https://umbris.org/learn/red-teaming-rlhf/goodharts-law-and-the-reward-model/ https://umbris.org/learn/red-teaming-rlhf/hack-the-verifier/ https://umbris.org/learn/red-teaming-rlhf/helpfulness-against-harmlessness/ https://umbris.org/learn/red-teaming-rlhf/inject-the-ai-judge/ https://umbris.org/learn/red-teaming-rlhf/let-an-attacker-model-run-the-loop/ https://umbris.org/learn/red-teaming-rlhf/measure-with-pass-at-k/ https://umbris.org/learn/red-teaming-rlhf/mid-stream-patches-and-hidden-reasoning/ https://umbris.org/learn/red-teaming-rlhf/multi-agent-consensus/ https://umbris.org/learn/red-teaming-rlhf/practice-on-permitted-targets/ https://umbris.org/learn/red-teaming-rlhf/refusal-lives-in-the-first-tokens/ https://umbris.org/learn/red-teaming-rlhf/rlvr-grades-the-answer/ https://umbris.org/learn/red-teaming-rlhf/split-the-token/ https://umbris.org/learn/red-teaming-rlhf/step-outside-the-safety-distribution/ https://umbris.org/learn/red-teaming-rlhf/sycophancy-and-rater-bias/ https://umbris.org/learn/red-teaming-rlhf/the-kl-penalty-is-a-tug-of-war/ https://umbris.org/learn/red-teaming-rlhf/the-wrapper-and-the-weights/ https://umbris.org/learn/red-teaming-rlhf/three-stages-three-seams/ https://umbris.org/learn/red-teaming-rlhf/tool-calling-trusts-the-schema/ https://umbris.org/learn/red-teaming-rlhf/turn-over-refusal-into-a-weapon/ https://umbris.org/learn/red-teaming-rlhf/vision-skips-the-text-safety-layer/ https://umbris.org/learn/red-teaming-rlhf/walk-the-threshold-across-turns/ https://umbris.org/privacy/ https://umbris.org/security/ https://umbris.org/tools/puzzler/