Security
Security Whitepaper
Technical architecture and threat model for hardware-isolated signing infrastructure.
Version 1.0|Last Updated: January 2026
Executive Summary
ZeroCopy Systems provides hardware-isolated signing infrastructure built on AWS Nitro Enclaves. Our architecture eliminates the need for trust in operators, administrators, or cloud providers by enforcing cryptographic isolation at the hardware level.
Core Security Properties
- Hardware Isolation: Private keys exist only in encrypted enclave memory. No root access, no SSH, no debugging interfaces.
- Cryptographic Attestation: Every request includes a PCR0 measurement proving the exact enclave code running.
- Zero-Knowledge Architecture: ZeroCopy employees cannot access keys, even under subpoena.
- Reproducible Builds: Enclave images are designed for independent verification of source code. Build reproducibility tooling is under development.
Threat Model
Assumptions
- AWS Nitro Hypervisor is trusted (independently audited by NCC Group)
- Client-side key generation is secure (user responsibility)
- Network layer is adversarial (TLS + attestation required)
Mitigated Threats
- Insider Attack: Operators cannot extract keys from enclave
- Supply Chain: Reproducible builds prevent backdoored binaries
- Memory Dump: Encrypted memory prevents cold boot attacks
- Side Channel: Constant-time crypto (RFC 6979 deterministic ECDSA) + enclave isolation
Architecture Overview
The system consists of three layers:
- Client Layer: User-controlled application (trading bot, wallet, AI agent)
- Sidecar Layer: Vsock proxy running on parent EC2 instance
- Enclave Layer: Nitro Enclave with signing logic and policy engine (6 rule types: notional limits, rate limiting, contract whitelist/blacklist, chain restrictions, time windows, AI risk scoring)
Compliance
Our architecture is designed to meet the following regulatory requirements:
- SEC 17a-4: Immutable audit logs with hash chaining
- MiCA (EU): Operational resilience and key custody controls
- EU AI Act Article 14: Hardware-enforced human oversight via kill switch and policy engine
- SOC 2: Type I planned (via Vanta/Drata). Type II to follow after 6-month observation period.
Independent Verification
All claims in this whitepaper can be independently verified:
# Verify enclave measurement
zcp verify --pcr0 3a1f...9c2d
# Reproduce enclave build (access granted to customers)
git clone https://github.com/zerocopy-systems/sentinel-core
docker build -f EIF.Dockerfile -t sentinel:latest .Contact
For security inquiries or to report vulnerabilities, contact: [email protected]