Threat Prevention Layer
Block prompt injections, jailbreak overrides, and outbound PII/IP leakage.
The Threat Prevention Layer acts as a bidirectional firewall for LLM payloads. It sanitizes inputs to prevent prompt injection and scrubs outputs to prevent data exfiltration and PII leakage.
Inbound Sanitization
User inputs are vector-matched against known adversarial fingerprints and zero-day topological structures indicating jailbreak attempts. Safe inputs are passed; malicious inputs trigger an anomaly flag.
Outbound Scrubbing
- PII Masking (SSN, Credit Cards, Emails)
- Proprietary IP stripping
- Toxicity classification
EAAP Protocol v1.0Verified Architecture