Skip to content

v1.0.0 - Zero-Trust & Quantum-Resistant Agent Governance

Choose a tag to compare

View all tags
@SRN0129 SRN0129 released this 25 Oct 22:04
· 55 commits to main since this release
v1.0.0
cab5e33

AgentTrust v1.0.0

Production Release: Zero-Trust Security Gateway, Cryptographic Audit Ledger, and Post-Quantum Governance for Autonomous AI Agents

AgentTrust provides an interoperable, vendor-agnostic security control plane designed to enforce capability boundaries, protect against indirect prompt injections and SSRF, manage human approvals with single-use replay protection, and cryptographically notarize agent operations.

What's Included in v1.0.0

1. Zero-Trust Policy Engine & Capability Attenuation

  • Deny-by-default execution policy across arbitrary tools and resources.
  • Support for prefix, suffix, and infix glob patterns (document:*, *:read, cloud:*:deploy).
  • Cascading delegation protocol: parent orchestrators can spawn subagents with strictly downscoped permissions.

2. Semantic Blast Radius Scoring & Human-in-the-Loop Review

  • Multi-factor blast radius analyzer evaluating action reversibility, sensitive target exposure, and financial liability ($[0.0, 1.0]$).
  • Automated escalation to human-in-the-loop review for high-impact actions.
  • 1-Click decision callback URLs for Slack, Microsoft Teams, and email.
  • Single-use cryptographic execution nonces with atomic burn (CONSUMED), providing strict HTTP 409 replay attack defense.

3. Verifiable Audit Ledger & Zero-Knowledge Merkle Proofs

  • Append-only backward hash chaining linking every event to the prior ledger entry ($O(1)$ cryptographic tamper detection).
  • Periodic digital notarization using asymmetric ECDSA P-256 signatures.
  • Binary Merkle tree generation providing self-contained, air-gapped inclusion proofs for offline auditor verification without database disclosure.

4. Transparent Egress Proxy & SSRF Defense Shield

  • Transparent proxy intercepting outbound HTTP agent requests.
  • Strict network isolation dropping calls to loopback interfaces (localhost, 127.0.0.1, ::1), private RFC1918 subnets (10.0.0.0/8, 192.168.0.0/16), and cloud instance metadata services (169.254.169.254, metadata.google.internal).

5. Goal Drift & Prompt Injection Guard

  • Real-time semantic analysis comparing an agent's declared mission against proposed tool arguments.
  • Detection and neutralization of indirect prompt injections and exfiltration sinks pre-flight.

6. Process-Isolated Sandbox & Emergency Air-Gap Killswitch

  • Isolated execution sandbox with prototype pollution guards and hard 3000ms watchdog timeouts.
  • Global air-gap killswitch revoking all active credentials and pending nonces across all agents in $< 15\text{ ms}$.

7. Quantum AI Circuit Governance & NIST Post-Quantum Cryptography (PQC)

  • Pre-flight OpenQASM parsing enforcing hardware qubit boundaries ($\le 127$ qubits) and coherence depth limits ($\le 500$).
  • NIST FIPS 204 (ML-DSA / Dilithium) Level-5 hybrid post-quantum attestation envelopes defending historical audit trails against "Harvest Now, Decrypt Later" quantum threats.
  • Quantum shot micro-budgeting preventing QPU queue jamming and runaway financial burn.

8. Verification & Test Suite

  • Comprehensive suite of 70 automated tests across 12 test suites passing locally and in continuous GitHub Cloud CI.

Package Distribution

  • npm / GitHub Packages: @srn0129/agenttrust@1.0.0
  • Release Asset: srn0129-agenttrust-1.0.0.tgz (packed npm distribution with compiled server and client)
Assets 3
Loading