# CAIN42 — ARGUS Patent Collision Analysis (Prompt 14)

> **Not legal advice. No patent filed. No freedom-to-operate opinion.** Colours are a self-assessment of overlap
> risk for engineering triage only. A professional prior-art search and counsel review are required before any
> proprietary claim.

| Mechanism | Colour | Note |
|---|---|---|
| Continuous assurance state (composite runtime primitive) | GREEN | runtime governance is emerging (AgentBound, ARG; 2026) but a signed assurance object covering behavior+environment+world+effect is not a known product category |
| Autonomous assurance lease (no self-renewal) | GREEN | token TTL/expiry exists; the "authority is a lease the agent cannot renew" formulation is differentiated |
| Behavioral envelope vs actual execution path | YELLOW | AgentBound (2026) composes delegated authorization, behavioral constitutions and site action contracts; meaningful overlap requires counsel |
| Path-aware composition invariance (salami) | YELLOW | cumulative resource limits exist; authority-layer composition of individually-permitted actions less common |
| Continuous re-authorization ladder | YELLOW | risk-adaptive authz / step-up auth exist; the material-change -> impact -> action ladder for agents differs |
| Independent verifier mesh (structural independence) | GREEN | quorum/oracle independence is old (blockchain); applying it to agent assurance is differentiated |
| Memory-to-authority firewall | YELLOW | memory poisoning / provenance research is active (governed memory); the authority-ceiling formulation differs |
| Assurance certificate + clean-room verify | GREEN | attestation/certificates are broad; assurance-state certificate is less crowded |
| Evolution -> capability -> authority coupling | YELLOW | self-evolving-agent governance is an active 2026 research area |
| World-model authority assurance | ORANGE | world-model / sim-to-real authority gates are a broad robotics/safety field |
| Physical-AI assurance adapter | ORANGE | robotics/AV safety middleware is crowded; this is SIMULATED |
| Adversarial assurance lab | GREEN | deterministic assurance lab is engineering, not a patent target |
| CAAP open protocol | GREEN (open) | position as a proposed open protocol, not a proprietary claim |

**Rule:** no item is claimed as patented or novel. ORANGE requires counsel review before any proprietary claim.
