DEV Community

Scott
Scott

Posted on

One line to add post-quantum audit trails to any Python AI agent

I'm a solo founder building compliance infrastructure for AI agents.
This is the core problem I'm solving.

If you're building AI agents, you have a logging problem
you probably haven't thought about yet.

Not application logs. Compliance logs — cryptographically
signed, tamper-proof records of what your agent decided,
when, and why. The kind that hold up in a regulatory audit.

The EU AI Act (Article 12) mandates exactly this for
high-risk AI systems. And most agent frameworks produce
nothing close to it.

I built Rubric Protocol to fix that.


What it does

Every agent action gets:

  • A post-quantum signature (ML-DSA-65)
  • A Merkle-aggregated attestation bundle
  • An anchor written to Hedera Consensus Service mainnet
  • A W3C Verifiable Credential you can export

That's a tamper-proof, court-admissible audit trail for
every decision your agent makes.


Install

pip install autogen-rubric


Instrument your agent in one line

import rubric
rubric.instrument()

That's it. Works with:

  • LangChain
  • LangGraph
  • CrewAI
  • AutoGen
  • OpenAI Agents SDK
  • Google ADK
  • Haystack
  • Semantic Kernel
  • Pydantic AI
  • DSPy
  • LlamaIndex
  • Strands / Bedrock AgentCore
  • and more

Why post-quantum?

Current ECDSA signatures will be breakable by quantum
computers within the decade. If your audit trail gets
subpoenaed in 2031, you want signatures that still hold.

ML-DSA-65 (CRYSTALS-Dilithium) is NIST-standardized.
We implemented it natively via a C++ N-API addon — 52x
faster than pure JS equivalents.


Architecture in 30 seconds

  1. Agent action fires
  2. Rubric signs it with ML-DSA-65
  3. Actions batch into a Merkle tree (1,000,000:1 compression)
  4. Root hash anchors to Hedera mainnet via HCS
  5. W3C VC issued against that anchor

Load tested at 3,247 RPS, 0.13% error rate.
First mainnet VC issued April 3, 2026.


Try it

pip install autogen-rubric

Docs + free developer key: rubric-protocol.com

Happy to answer questions on the ZK layer, the Merkle
architecture, or the EU AI Act compliance angle.

Top comments (0)