Security events

Prompt-injection defense layer hits. 7-layer guard chain.

Critical
1
Error
1
Warnings
3
Blocked rate
57%
Severity
Layer
7-layer defense chainInbound message → each layer can block or annotate
  1. 1Regex block
    Bans known injection phrases before LLM sees them.
    hits1
  2. 2Prompt template
    Hard-coded SCOPE prefix wraps every user message.
    hits1
  3. 3Claude extract
    Confidence + schema validation on Sonnet 4.5 output.
    hits1
  4. 4Post-validate
    Zod schema + business-rule checks after extraction.
    hits1
  5. 5SEND filter
    Single chokepoint — opt-out, 24h, block-rate, signature.
    hits1
  6. 6Rate limit
    Per-tenant + per-conversation message-rate ceiling.
    hits1
  7. 7T&C override
    Blocks unsanctioned terms-and-conditions edits.
    hits1