diff --git a/docs/static/layered-security.svg b/docs/static/layered-security.svg new file mode 100644 index 0000000..1c8b2eb --- /dev/null +++ b/docs/static/layered-security.svg @@ -0,0 +1,65 @@ + + + + + + + + + + + + + + + + + + + + + + + LAYER 4 — RUNTIME CONTROLS + Per-tool guardrails: HITL, PITL, AITL, Deny, Filter, Content Safety + + + + LAYER 3 — METAPROMPT / SYSTEM MESSAGE + Behavioral boundaries, persona constraints, output formatting rules + + + + LAYER 2 — PLATFORM SAFETY + Azure AI Content Safety, Prompt Shields, input/output filtering + + + + LAYER 1 — MODEL + Built-in safety training, RLHF alignment, refusal behaviors + + + + claude-opus-4.6 + Tier 3 • Safe + + + gpt-4.1 + Tier 2 • Standard + + + gpt-5-mini + Tier 1 • Cautious + + + + Strongest + + + Outermost + + + + + Defense in Depth +