Contributing to Cognitive Insight

Building a Verifiable Future Together
License: Apache License 2.0

Why Open Source

When an AI system needs to be audited - whether for legal, regulatory, operational, or ethical reasons - people should be able to examine the evidence behind how it was designed, controlled, and evaluated. Open access makes it possible to inspect the “receipts”: the policies, controls, records, and technical artifacts that demonstrate how a system operates and whether it can be trusted.

This resource may not, on its own, satisfy every requirement in every law, regulation, or industry standard. But it is intended to provide a transparent foundation that organizations, auditors, researchers, and practitioners can build upon.

My goal is larger than releasing tools or documentation. I hope to help foster an independent community - separate from the interests of corporations, governments, or any single regulatory body - that can evaluate AI systems openly and rigorously. AI should be trusted not because an institution says it is safe, but because its controls, decisions, and evidence can be examined.

That is why I am open sourcing this work - to help make AI more transparent, governable, controllable, and demonstrably aligned with the best interests of humanity.

🗺️ Open-Source Collaboration Roadmap

To turn this vision into a living, breathing reality, we are rallying the global open-source community around four critical engineering workgroups:

1. Workgroup: API & SDK Standards (Thin, Fast, and Agnostic)

To gain widespread adoption, the evidence layer must have a near-zero latency impact on running systems. We need to build thin, language-agnostic client libraries that wrap model calls and tool executions.

The Mission: Design and implement client SDKs in Python, Go, Rust, and TypeScript that intercept runtime events, apply deterministic canonicalization (RFC 8785 JCS), and generate lightweight micro-receipts.

Key Projects:

  • Pre-Action Interceptors: Middleware wrappers for popular agentic frameworks (such as LangChain, AutoGen, and Semantic Kernel) that automatically generate and verify pre_action_proof_bundles before a tool is executed.
  • Lightweight Cryptographic Envelopes: High-efficiency signing modules that can generate both classical (Ed25519) and post-quantum (ML-DSA-65 / Falcon) signatures without adding runtime overhead.
  • JSON Canonicalization Engines: Optimized standard library extensions that enforce strict UTF-8 NFC string normalization, float serialization, and key sorting.

2. Workgroup: Open-Source Verification Tooling (CLI & Library)

Trust must never require accessing a live corporate database. Anyone holding an exported "Audit Pack" should be able to verify its validity locally, on their own machine, with zero third-party dependencies.

The Mission: Develop a robust CLI tool and verification library that parses standardized audit pack manifests and proves their absolute cryptographic integrity.

Key Projects:

  • ci-verify CLI: A downloadable command-line tool written in Rust (for speed and security) that takes an audit_pack.json and returns a definitive verification report.
  • Merkle Path Validator: A standalone utility that takes a specific receipt, a Merkle proof, and a target batch root, and mathematically proves membership without revealing adjacent transaction data.
  • Chain-Integrity Analyzer: A validation engine that walks the back-linked hashes of a transaction sequence (using predecessor_hash fields) to detect if any historical record has been altered, injected, or deleted.

3. Workgroup: Dynamic Control Mappings & Policy Registers

Compliance shouldn't be an afterthought. We need to translate high-level legal frameworks into machine-evaluable policy files that can actively block dangerous actions at runtime.

The Mission: Maintain an open registry of machine-evaluable YAML/JSON policy sets that map operational audit receipts directly to global regulations.

Key Projects:

  • EU AI Act Crosswalk: Codifying the logging, traceability, and human oversight requirements of Article 12/14 into standard policy_rules and gate_definitions.
  • NIST AI RMF & ISO/IEC 42001 Pack: Pre-configured rule templates that verify data lineage, training boundaries, bias evaluations, and continuous system monitoring.
  • Regulatory API Connectors: Automated scrapers and sync-scripts that connect to public policy registers (like the OECD Policy Database) and generate pull requests to update active policy rules when legal requirements shift.

4. Workgroup: Post-Quantum Security & Zero-Knowledge Proofs (ZKP)

For audit trails to remain legally binding decades into the future, they must survive both the quantum transition and strict privacy scrutiny.

The Mission: Ensure long-term, privacy-preserving cryptographic custody.

Key Projects:

  • Hybrid Dual-Signature Envelopes: Production testing of hybrid classical/quantum signature boundaries to verify interoperability across legacy enterprise key management systems.
  • Zero-Knowledge Receipt Disclosures: Implementing ZK-SNARK protocols to allow an operator to prove to an auditor that "this agent executed tool X under a valid approval from authorized user Y" without revealing the actual identities of the user, agent, or sensitive parameters in the transaction.

🤝 Join Us: Let's Build a Verifiable Future

We believe that AI is too powerful to be governed by promises, NDAs, or opaque algorithms. The only way to ensure that autonomous systems act in the best interest of humanity is to make their operational boundaries completely transparent and mathematically verifiable.

If you are a developer, security engineer, compliance expert, or open-source advocate, your contribution is needed:

  • Fork the Repository: Grab the schema contract and architecture guides, and build a prototype.
  • Submit a Pull Request (PR): Help us refine the 60-table database model, write an SDK client, or add a regulatory mapping.
  • Deploy a Node: Implement the reference architecture inside your organization and share your operational feedback.
  • Spread the Word: Help us make cognitiveinsight.ai the global, independent home for cryptographically verifiable AI accountability.

Let's build a verifiable future for AI together.