Most AI governance systems check approval once. Then they assume the agent is still authorised to act.
That assumption fails silently.
Policy changes. Delegation revokes. Evidence expires. Yet the agent continues executing under a stale authority context.
I built Nanogate – a software‑only gate that re‑tests admissibility before every action, in ~530 nanoseconds.
It answers the question that most governance tools ignore:
“Does this agent still deserve to execute right now?”
The Problem with Point‑in‑Time Approval
A typical AI governance flow looks like this:
Approval – a human or policy engine says “yes” at time T₀.
Execution – the agent acts at time T₁ (seconds, minutes, or days later).
Between T₀ and T₁, many things can change:
The policy version is updated.
The delegation chain is modified.
The agent’s identity or session mutates.
Supporting evidence expires.
A malicious actor replays an old approval.
Traditional systems log these changes but do not stop the agent. The result: an action that was approved but is no longer admissible at execution time.
Approval is not enough.
Continuous Admissibility
I propose a different principle: every action must re‑prove its admissibility immediately before execution.
The agent must present:
Its stable identity (agent_id, session_id, memory state)
The active reference frame (policy version, delegation chain, external state hash)
A monotonic timestamp and a nonce (to prevent replay)
The gate then:
Hashes the identity and reference frame using xxHash64 (fast, non‑cryptographic)
Compares the hashes with the last verified state
If unchanged and timestamp increased → ADMIT
Else → DENY with a clear reason (identity drift, policy drift, etc.)
Emits a BLAKE3 proof hash of all inputs (signed, replayable, court‑admissible)
This is Continuous Admissibility – a category I am defining and implementing.
Nanogate: A Reference Implementation
Nanogate is a Rust library and CLI that implements the gate. It is:
Fast – median 530 ns per evaluation (Criterion benchmark)
Deterministic – the same input always produces the same output
Adversarially validated – 0 false admits after 100k random mutations
Reliable – 0 false denies after 100k stable continuity traces
Lightweight – no hardware attestation, no external dependencies beyond Rust std
Performance
bash
$ cargo bench
nanogate evaluate time: [528.91 ns 530.01 ns 531.18 ns] That’s ~1.9 million evaluations per second per CPU core.
Faster than the time light travels in 160 metres.
Correctness Validation
Test Type Cases Result
Unit tests 4 ✅ pass
Property tests (stable context, drift, timestamp) 4 ✅ pass
Adversarial mutation (false admits) 100,000 ✅ 0 false admits
Stable continuity (false denies) 100,000 ✅ 0 false denies
Run the full suite yourself:
bash
git clone https://github.com/a1k7/nanogate cd nanogate
cargo test --release Why Rust?
No runtime overhead – the hot path avoids allocations, JSON parsing, and interpreted code.
xxHash64 is ~10x faster than SHA‑256 for non‑cryptographic hashing.
BLAKE3 is hardware‑accelerated on modern CPUs (AVX‑512, SSE) and still very fast.
pyo3 bindings exist if you need to call Nanogate from Python (optional).
Next Steps: The Continuous Admissibility Protocol (CAP) Nanogate is not the end goal. It is the reference implementation of a larger idea.
I am drafting CAP – the Continuous Admissibility Protocol – a lightweight open standard for runtime admissibility proofs.
Every CAP‑compliant agent would emit a proof containing:
agent_id
observer_hash (identity + session + memory)
constitution_hash
policy_hash
continuity_hash (chained from the previous proof)
admissible (boolean) No vendor lock‑in. No black boxes.
The Runtime Governance Index will benchmark agent frameworks (LangGraph, CrewAI, AutoGen, etc.) for CAP compliance. Public leaderboard. Transparent criteria.
Commercial Licensing
Nanogate is open source under MIT / Apache‑2.0 for non‑commercial and internal use.
For embedding Nanogate inside proprietary agent runtimes, a commercial license is required: Perpetual use in one product
Email support for one year
$5,000 one‑time fee + $1,000/year support renewal
Contact: [akhilesh@decisionassure.io](mailto:akhilesh@decisionassure.io)
Try It Yourself
bash
git clone https://github.com/a1k7/nanogate cd nanogate
cargo build --release
cargo run --release
Final Thought
The AI governance community has built many tools for approval.
What we lack is a tool for continuous admissibility – proof that an agent still deserves to act at the exact moment of execution.
Nanogate is my contribution to that gap.
Approval is not enough. Continuity first.
If you are building agent frameworks, runtime governance systems, or compliance tooling – I invite you to read the CAP spec (coming soon) and run the Nanogate benchmark. Open source is free. Commercial licensing is available. Let’s make runtime governability the new standard. #rust #aigovernance #runtime #continuousadmissibility.
Continuity first.