JetStream Announces Clearance, a Reasoning Engine That Denies Agent Actions Against an Approved Design
JetStream Security announced Clearance on 2026-09-02, a zero-trust reasoning engine sitting at its AI Gateway that maps every request to the agent or user making it, the approved design authorizing it, the tools being invoked and what the specific call is about to do, then denies dangerous sequences such as exfiltration patterns before they execute rather than logging them afterward. The company launched with a $34M seed round, is demoing Clearance at Fal.Con this week, and was named in CrowdStrike's new AI Partner Specialization. The design point worth stealing is that authorization is evaluated against a declared agent design, not a static permission list, which is how you catch a sequence of individually legal tool calls.
↳ Follow the thread