
The P9 MCP Governance Proxy establishes that safety can be enforced protocol-transparently for any MCP-compatible agent — but at a cost: when a tool call triggers a governance HALT, the suspended tool call blocks until a human returns a signed Accountability Proof Block (APB). This design sacrifices liveness for safety under inevitable human response delays. We resolve this tension by introducing escrow-based non-blocking governance. Non-blocking refers to continued agent progress: the suspended tool call remains in escrow while the agent session continues. When a risky tool call triggers a HALT, its state is serialised into a persistent escrow entry and deposited in a priority queue rather than blocking the agent pipeline. We formalise three results: T10.1 (Non-Blocking Soundness) — no risky tool call executes without a valid APB satisfying predicates V1–V6, with at-most-once execution semantics; T10.2 (Timeout Consistency) — fallback decisions are locally equivalent to explicit governance decisions; T10.3 (Escrow Liveness) — the first liveness theorem in the series, establishing that if the human signs within the timeout window, the suspended call resumes with bounded latency. Six experiments validate the construction: P10 achieves up to 81× higher throughput than P9's blocking model at 80% halt rate; escrow overhead is negligible at P95 = 6.4 µs. Paper 10 of the Agent Governance Series.
LLM agents, agent governance, human oversight, non-blocking, escrow, asynchronous authorization, liveness, safety, MCP, accountability proof block, AI safety, runtime verification
LLM agents, agent governance, human oversight, non-blocking, escrow, asynchronous authorization, liveness, safety, MCP, accountability proof block, AI safety, runtime verification
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 0 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Average | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Average | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Average |
