The Exfiltration Gate

In the SLM entitlement build, a local Phi-3 — swappable to any sector-specific persona — scores exfil intent on-host. Your PII never leaves the server.

Every suspect ioctl gets two passes. Stage one is a kernel-side heuristic on the hot path: size estimate plus reservoir sample, sub-microsecond. Stage two — in the SLM entitlement build — hands the survivors to the active on-host persona model, which scores intent on process and transfer metadata (not the transferred bytes) and returns a 0–100 risk rating with a reason; the default install ships without the model and runs the heuristic alone. The persona's threshold — set by your operator, not by us — decides the outcome: allow, alert, or terminate the offending process. The model runs on the same host with no cloud round-trip and no third-party API. The Persona Switchboard hot-swaps the cartridge at the driver boundary with zero downtime.