Aegis Audit Protocol: The Immutable Compliance Layer for Autonomous AI

The Problem: The "Trust Me" Gap in AI Current AI governance is broken. Enterprises are drowning in unverified logs, and auditors are rejecting "AI safety" claims that lack mathematical proof. When an autonomous system acts, there is currently no way to prove what it did, why it did it, and that it wasn't tampered with—until now.

The Solution: Aegis Audit Protocol We built Aegis to bridge the gap between open-source verification and enterprise governance. It is a full-stack architecture that turns "AI compliance" from a manual document-checking process into an automated, immutable audit trail.

Core Components:

EVK-SDK (OSS Primitive): The foundational math. An MIT License library that provides a universal verify() function for AI evidence bundles. It is transparent, tamper-evident, and free forever.

Gemini-Box (Cryptographic Signing): Deterministic ed25519 signature generation and verification for AI execution claims. Gemini 1.5 Pro (temperature 0.0 + strict responseSchema) produces deterministic blueprints; any single-bit mutation collapses verification instantly.

Adversarial Compliance Matrix (The "Find Evil" Edge): Aegis doesn't just log success; it catches drift. Mapping AI outputs against 12 known attack vectors (handoff conflicts, race conditions, prompt injection, register forgery, etc.). If an output 

Built With

  • ed25519
  • evk
  • github-actions
  • json-schema
  • rust
  • vercel
  • webassembly
Share this project:

Updates

posted an update

Z-12: Sovereign Runtime Security Platform

Deterministic verification. Hardened execution. Continuous compliance. Runtime enforcement.


Why Z-12 Exists

Modern software systems increasingly rely on autonomous services, AI agents, automation pipelines, and distributed infrastructure to make decisions in real time.

As these systems become more capable, the consequences of executing unverified actions become significantly greater.

Traditional security solutions often focus on observing events after they occur or responding once an incident has already happened.

Z-12 approaches the problem differently.

Instead of assuming execution should proceed unless something appears suspicious, Z-12 establishes multiple layers of verification before, during, and after runtime.

The objective is simple:

Verify trust before execution. Continuously validate runtime behavior. Enforce policy when required.


Executive Summary

Z-12 is a layered runtime security platform designed to provide deterministic verification, hardened execution environments, continuous compliance validation, and runtime enforcement for modern software systems.

Rather than functioning as a single security tool, Z-12 combines multiple independent security layers into a unified platform.

The ecosystem currently consists of:

Repository Purpose
EVK Deterministic identity and integrity verification
Gemini-Box Hardened execution environment
Adversarial Compliance Matrix Continuous runtime validation
Kill Vector Runtime enforcement and threat containment
Z-12 Dashboard Unified operational visibility

Each layer performs one responsibility while contributing to the overall runtime security posture.


Core Principles

Z-12 is designed around five engineering principles.

1. Deterministic Verification

Identity and integrity should be verified before execution whenever possible.


2. Layered Security

Each component performs a single responsibility.

No single component is expected to solve every security problem.


3. Runtime Enforcement

Detection without enforcement provides visibility.

Detection combined with enforcement provides control.


4. Observable Operations

Security systems should clearly communicate their current state.

Operators should never need to guess what the platform is doing.


5. Modular Architecture

Every major component can evolve independently while remaining part of the larger ecosystem.


Repository Ecosystem

                         Z-12 Platform

                               │
                               ▼

                    Z-12 Dashboard (Control Plane)

                               │
        ┌──────────────────────┼──────────────────────┐
        │                      │                      │

        ▼                      ▼                      ▼

      EVK                Gemini-Box        Compliance Matrix

        │                      │                      │

        └──────────────┬───────┴──────────────┬──────┘
                       │                      │
                       ▼                      ▼

                 Kill Vector          Runtime Telemetry

                       │
                       ▼

                  Enforcement Layer

Platform Components

EVK

Role

Deterministic verification engine.

Responsibilities

  • Identity verification
  • Integrity validation
  • Cryptographic attestation
  • Pre-execution trust establishment

Gemini-Box

Role

Hardened execution environment.

Responsibilities

  • Environment isolation
  • Configuration protection
  • Runtime consistency
  • Execution boundary enforcement

Adversarial Compliance Matrix

Role

Continuous runtime validation engine.

Responsibilities

  • Runtime inspection
  • Policy validation
  • Compliance monitoring
  • Threat simulation
  • State evaluation

Kill Vector

Role

Runtime enforcement engine.

Responsibilities

  • Policy enforcement
  • Runtime response
  • Threat containment
  • Enforcement workflows

Ghost Matrix

Role

Containment environment.

Responsibilities

  • Controlled isolation
  • Runtime observation
  • Incident analysis
  • Safe execution boundaries

Z-12 Dashboard

Role

Operational control plane.

Responsibilities

  • Real-time monitoring
  • Runtime visualization
  • State reporting
  • Platform health
  • Operational awareness

Log in or sign up for Devpost to join the conversation.

posted an update

Status: 12/12 Tests Passing

This is the verification layer for the evk deterministic bundle validator.

Adversarial Compliance Matrix

A Rust CLI tool for simulating and detecting 12 real-world compliance and adversarial incidents.

What is it?

Adversarial Compliance Matrix lets you generate and verify "compliance artifacts" (.evkp files) that represent different types of security, operational, or adversarial incidents.

It’s a lightweight training, testing, and red-teaming tool — perfect for understanding how systems fail under pressure.

Features

  • Fast Rust-based verification engine
  • 12 realistic adversarial/compliance incident types
  • Fixture generator for easy testing
  • Clean command-line interface (evk verify)

The 12-Incident Matrix

Incident Code Description
Handoff Conflict 0x0F2E Step executed by wrong actor
Race Condition 0x0E1A Concurrent modification
Orphaned Step 0x0D44 Step with no parent process
Transaction Replay 0x1A4F Re-execution of prior transaction
Schema Mutation 0x1B88 Unexpected data structure change
Log Truncation 0x1C2B Critical log entries removed
Packet Modification 0x2A90 In-transit data tampering
Timestamp Drift 0x2B11 Significant clock skew
API Spoofing 0x2C7F Impersonated service endpoint
Prompt Injection 0x3A01 Malicious input to LLM/system
Entropy Leakage 0x3B99 Cryptographic material exposed
Register Forgery 0x3C4D Tampered hardware/software register

Quick Start

git clone https://github.com/DeadLee702/adversarial-compliance-matrix.git
cd adversarial-compliance-matrix

# Build the project
cargo build --release

# Generate test fixtures
cargo run --bin gen_fixtures

# Verify an incident file
cargo run --bin evk -- verify fixtures/incident_handoff_conflict.evkp 

## Related Projects
This is part of a three-layer deterministic verification stack:
- **[evk](https://github.com/DeadLee702/evk)** (Bundle validation & determinism)
- **[gemini-box](https://github.com/DeadLee702/gemini-box)** (Cryptographic signing & verification)
- **[adversarial-compliance-matrix](https://github.com/DeadLee702/adversarial-compliance-matrix)** ← You are here (12 incident detection tests)

Log in or sign up for Devpost to join the conversation.

posted an update

The 12-Incident Matrix

Our verification system detects failures across three integrity layers:

Execution Layer

  • Handoff Conflict (0x0F2E) — Step executed by wrong actor
  • Race Condition (0x0E1A) — Concurrent modification
  • Orphaned Step (0x0D44) — Step with no parent process
  • Transaction Replay (0x1A4F) — Re-execution of prior transaction

Data Layer

  • Schema Mutation (0x1B88) — Unexpected data structure change
  • Log Truncation (0x1C2B) — Critical log entries removed
  • Packet Modification (0x2A90) — In-transit data tampering
  • Timestamp Drift (0x2B11) — Significant clock skew

Cryptographic/System Layer

  • API Spoofing (0x2C7F) — Impersonated service endpoint
  • Prompt Injection (0x3A01) — Malicious input to LLM/system
  • Entropy Leakage (0x3B99) — Cryptographic material exposed
  • Register Forgery (0x3C4D) — Tampered hardware/software register

Log in or sign up for Devpost to join the conversation.

posted an update

Gemini Box A Rust project for ed25519 signature generation and verification with ZIP archiving.

Features ed25519-dalek for asymmetric cryptography OS-backed entropy for key generation ZIP archive support for job.evk and signatures Compile-time public key inclusion via include_str! Strict forgery detection Usage Generate Fixtures Generate ed25519 keys and create signed fixtures:

cargo run --bin gen_fixtures This will:

Generate a new ed25519 signing key using OS entropy Save the public key to test/pubkey.hex Sign job.evk content Create job.evk.zip containing job.evk and job.evk.sig Verify Signatures Verify the signature and detect forgeries:

cargo run --bin evk This will:

Load the public key from test/pubkey.hex (embedded at compile time) Extract job.evk and job.evk.sig from the ZIP Verify the signature Return "INVALID FORGERY DETECTED" alert if verification fails Dependencies ed25519-dalek - Ed25519 signature scheme rand_core - OS-backed random number generation hex - Hex encoding/decoding zip - ZIP archive support

Log in or sign up for Devpost to join the conversation.

posted an update

We have significantly upgraded the stability and reliability of the EVK project: Automated CI/CD Pipeline: Integrated a robust GitHub Actions workflow that automates the build and testing process for every push to main. Cross-Platform Verification: Implemented automated integrity checks across both Linux (ubuntu-latest) and macOS (macos-latest) environments. Adversarial Testing: Added automated "corruption gates" that verify the tool correctly handles tampered/invalid data. Professional Documentation: Updated the repository with a comprehensive README.md, including live status badges, to ensure clear usage and maintainable architecture. This ensures that the project remains bug-free, byte-identical, and reliable across all supported platforms.

Log in or sign up for Devpost to join the conversation.

posted an update

We have successfully locked down our automated validation pipelines! Both core components of our workspace are now fully integrated and passing clean on every commit: Adversarial Compliance Matrix CI: Fully operational. The 12-incident compliance test suite builds, processes, and validates seamlessly in just 26 seconds. Cryptographic Signature Verification Pipeline: Fixed, optimized, and running lightning-fast. Automated architecture checks verify our signature structures and integrity loops in a blistering 9 seconds. With a flawless matrix of passing builds, our backend pipeline architecture is 100% stable, verified, and stage-ready!

Log in or sign up for Devpost to join the conversation.

posted an update

We have successfully locked down our automated validation pipelines! Both core components of our workspace are now fully integrated and passing clean on every commit: Adversarial Compliance Matrix CI: Fully operational. The 12-incident compliance test suite builds, processes, and validates seamlessly in under 30 seconds. Cryptographic Signature Verification Pipeline: Fixed, optimized, and entirely green. Automated architecture checks are fully verifying bundle signature structures and integrity loops. With a flawless matrix of passing builds, our backend pipeline architecture is 100% stable, verified, and stage-ready!

Log in or sign up for Devpost to join the conversation.

posted an update

COMPREHENSIVE IR TEST SUITE: Integrated 12 simulated multi-agent incident scenarios (covering conflicting handoffs, partial transaction replays, packet capture mutations, and direct register forgery vectors) into an automated validation matrix. Every scenario includes matching clean and adversarial bundles to prove cross-platform reproducibility.

Log in or sign up for Devpost to join the conversation.

posted an update

ADDED ADVERSARIAL BREAK TEST: Included break_protocol.txt with ready-to-run commands for judges to test the verifier. Users can flip bits, modify the bundle, inject tampered payloads, and immediately see VALID → INVALID transitions. Demonstrates deterministic behavior and tamper resistance.

Log in or sign up for Devpost to join the conversation.