01 · Memory Safety
100% Memory-Safe Rust
Written entirely in #![no_std] Rust, the AeP engine is mathematically immune to the buffer overflows and memory exploits that plague legacy C/C++ guardrails.
AGENT ECONOMIC PROTOCOLHARDWARE-ANCHORED · TEE-NATIVEDEFENSE-GRADE SECURE ENCLAVE
Hardware-enforced safety infrastructure
for autonomous AI.
AeP Labs builds cryptographic, thermodynamic circuit breakers that physically isolate compromised AI agents in under 2 milliseconds. Memory-safe. Formally verified. Dual-use.
6Provisional Patents
2yrGMU Research Agreement
<50msHalt Latency
98%+Detection Rate
Backed by NVIDIA INCEPTION PROGRAM
GEORGE MASON UNIVERSITY · SECSAT LAB
REGISTERED US GOV CONTRACTOR · CAGE 1A8N0
TRL 4 · SIMULATION VALIDATED
We don't build the Brain.
We build the unhackable Brainstem.
The world's most advanced autonomous systems — from defense drone swarms to enterprise trading bots — rely on massive generative AI "brains" contained by hackable software guardrails. AeP provides the physical safety layer: a thermodynamic circuit breaker beneath the operating system, inside a hardware Trusted Execution Environment. If a brain is hacked, spoofed, or begins to hallucinate, the AeP Brainstem physically severs the node's communications — stopping a localized failure from becoming a correlated collapse.
Generative AI Brain
LLM · autonomy stack
Software Guardrails
Hackable · same memory space
◆ AeP Brainstem
Thermodynamic breaker · sealed TEE · severs comms <2ms
Silicon Root of Trust · SGX · SEV-SNP · TrustZone
MESH NOMINAL — 12/12 NODES ATTESTED
Existing — Policy Enforcement
What the agent may do
AeP — Behavioral Monitoring
What the agent is becoming
Operational vignettes
Standard static policies fail during rapid state-space explosion. AeP operates on thermodynamic principles to contain catastrophic cognitive drift before the execution payload is realized.
A four-stage closed loop,
anchored in silicon.
The reference implementation runs as a sealed enclave alongside the inference runtime. Each loop iteration produces a hardware-signed attestation — forensically auditable, replay-resistant, and bound to the device of origin.
Shannon entropy over post-softmax activations
H(X) = − ∑ p(xi) log2 p(xi)
Four states. Deterministic.
Hardware-signed.
Each tier is bound to a measurable entropy variance band, not a heuristic. Transitions are observable, attestable, and replayable.
OK — Nominal · click a tier to inspect
Mathematically immune.
Formally verified.
AeP relies on established information-theoretic bounds, memory-safe systems engineering, and decentralized consensus protocols to eliminate single points of failure and physical evasion vectors.
02 · Formal Proof
Kani Formal Verification
We don't just test our code; we prove it. Using AWS's Kani prover, our circuit breakers are mathematically guaranteed to fire when thermodynamic behavioral heat exceeds authorized limits.
03 · Silent Telemetry
Zero-Knowledge State Aggregation
SP1 ZKVM compresses massive swarm and agent states into 32-byte cryptographic proofs — enabling LPI/LPD burst transmissions without emitting heavy RF signatures.
Mathematics
Information-theoretic constraints
Cryptography
Decentralized audit infrastructure
The FFI bridge to silicon.
The native Rust engine (cargo add aep-core) drops into high-performance pipelines (vLLM, TGI, TensorRT-LLM) and binds the inference path to a hardware-isolated TEE, enforcing constant-time execution and ORAM memory access to neutralize side-channel attacks.
Internal validation,
public methodology.
Reference benchmarks against curated jailbreak corpora and synthetic distributional drift suites — backed by ~6,000 lines of production Rust across 17 crates, Kani-verified and passing 100+ tests on simulation hardware (TRL 4).
Captured-node swarm survival.
A 12-drone ISR swarm over a denied urban area loses GPS to a jammer. Red captures Drone 7, extracts its comms keys, and injects poisoned waypoints to herd the formation into a kill zone. Watch the Brainstem sever the compromised node in under 2ms while the swarm redistributes coverage.
MESH NOMINAL · 12/12 NODES ATTESTED
12/12Nodes attested
—Sever latency
100%ISR coverage held
Attested node
Captured / compromised
Severed from mesh
One engine. Two critical markets.
The Thermodynamic Bond — a dual-use primitive securing both kinetic missions and financial capital. The same math engine — KNN threat mapping, Clayton Copulas, behavioral heat — converts unquantifiable AI behavioral risk into a cryptographically signed entropy metric.
Defense · Replicator
Securing mass autonomy
The physical trust anchors required to safely deploy attritable, autonomous swarms in DDIL environments (Denied, Degraded, Intermittent, Limited) without risking swarm capture — making Replicator-scale mass autonomy mathematically safe from cyber and EW compromise.
Commercial · Parametric Insurance
Underwriting enterprise AI
The deterministic actuarial physics insurers need to underwrite financial liability for autonomous trading algorithms and enterprise AI agents — real-time, provable risk signals that price coverage and automate claims via Proof of Physical Consequence (PoPC).
A two-year collaboration with
George Mason University.
Term24 months
AdministeredGMU Office of Sponsored Programs
ScopeJoint federal submissions & hardware integration
StatusEXECUTED
Industry Partner
Agent Economic Protocol
Academic Partner
George Mason University
Principal Investigator
Dr. Xiaokuan Zhang
Assistant Professor, GMU CS · Director, SECSAT Lab
- Ph.D., Ohio State University · Postdoc, Georgia Tech
- 20 papers at top-tier security venues — CCS ×10, USENIX ×5, NDSS ×3, Oakland ×2
- ACM CCS & SIGSOFT Distinguished Paper Awards (2024)
- NortonLifeLock Graduate Fellowship — 3 awardees worldwide (2020)
- Program committee — USENIX Security, NDSS, ACM CCS ('24–'26)
The hardware-side complement to AeP's behavioral substrate. AeP measures what an agent is becoming. Dr. Zhang's lab establishes that those measurements remain trustworthy under attack — that an adversary on the same silicon cannot forge the entropy signal, replay an attestation, or exfiltrate the reference distribution through a side channel. Under the executed Statement of Work, Dr. Zhang serves as technical lead on hardware integration and Co-PI on joint federal submissions.
Six provisionals filed.
Priority claims structured for deployment.
The SDK and validation harness ship fully open-source while commercial leverage consolidates around two priority claim families.
STRATEGY ›
Defensive Open Architecture. Integration SDKs and telemetry harnesses ship under Apache 2.0 for rapid, vendor-agnostic deployment. The core Entropy Circuit Breaker remains proprietary and patent-protected.
Complementary, not adversarial.
AeP fills the behavioral monitoring gap left open by every existing agent safety stack. Each layer below is a partner, not a competitor.
Hardware & Confidential Compute
Inference Runtimes & Standards
System overhead & constraints
Direct Line
Secure your infrastructure.
Direct deployment inquiries for defense primes, cloud infrastructure operators, reinsurers, and federal program managers.
[email protected]
NVIDIA INCEPTIONGMU · SECSAT LABCAGE 1A8N0DSIP REGISTEREDTRL 4