Key Features
- Prompt injection & jailbreak detection (direct + indirect)
- PII / sensitive data detection & masking (email, phone, SSN, credit card, custom patterns)
- Secrets leakage prevention (API keys, tokens, connection strings)
- Output scanning for harmful / off-policy content
- Powered by guard models (LlamaGuard-style) + rule engines — sub-100ms latency
- Configurable actions: block, mask, log, custom reply to agent
Outcomes
- Blocks 90%+ of common prompt-injection attacks before they reach tools
- Prevents accidental or malicious PII exfiltration
- Gives security teams visibility into attempted exploits