Security Policy

Security is critical for any protocol that orchestrates multiple AI models. This page covers how to report vulnerabilities, the security considerations you should be aware of when deploying ACP, and a production readiness checklist.

Supported Versions

Security patches are only provided for currently supported versions. Upgrade to the latest supported version to receive timely fixes.

Version	Supported
`4.0.x`	Yes
`< 4.0`	No

Reporting a Vulnerability

Do not open public issues

Never report security vulnerabilities through public GitHub issues, discussions, or pull requests. Use the private reporting channel described below.

How to Report

Send vulnerability reports by email. Include as much detail as possible to help us reproduce and assess the issue.

Detail	Value
Email	security@axiomprotocol.org
Expected Response Time	Within 48 hours

Your report should include the following information:

Description of the vulnerability
Steps to reproduce the issue
Potential impact and severity assessment
Suggested fix, if you have one

What to Expect

After you submit a report, the process follows a structured timeline:

Step	Timeline	Description
1. Acknowledgment	Within 48 hours	We confirm receipt of your report and assign a tracking ID.
2. Investigation	Days 3-30	We investigate the vulnerability and provide an estimated timeline for resolution.
3. Fix	During investigation	A fix is developed, reviewed, and tested internally.
4. Disclosure	Day 30	Coordinated disclosure with the reporter. We work with you on timing.
5. Credit	At disclosure	You are credited in the security advisory, unless you prefer to remain anonymous.

Security Considerations

The following areas require particular attention when deploying and operating ACP.

API Keys

Never commit API keys to version control
The .env file is gitignored by default -- use it for local development secrets
Use environment variables for all secrets in production
Rotate keys regularly and immediately after any suspected exposure

Worker API Authentication

Authentication is enforced via the x-openrouter-key header
Rate limiting is enforced at the edge: 100 requests per minute, 10 requests per second burst
Query text is not logged to protect user privacy

Python Backend

Run behind a reverse proxy (nginx, Caddy) in production -- never expose the FastAPI server directly
Enable HTTPS/TLS for all connections
Set the ACP_API_KEY environment variable for API authentication
Configure firewall rules to restrict access to trusted sources

Database Security

PostgreSQL connections must require authentication
Use strong passwords with a minimum of 16 characters
Enable SSL for all database connections
Perform regular encrypted backups and test restoration

Known Limitations

LLM Output Safety

ACP consensus does not guarantee safe or ethical output. Models can produce harmful content even when consensus is reached. For production deployments, implement content filtering and output validation as an additional layer on top of ACP.

Cost Controls

Consensus queries involve multiple LLM calls across multiple iterations and can become expensive, especially with large models. Set spending limits on your LLM provider account and use smaller models for testing and development.

Denial of Service

Rate limiting is enforced at the Cloudflare edge layer. If you are self-hosting the Python backend, implement additional rate limiting at the reverse proxy or application level to prevent abuse.

Production Checklist

Before deploying to production

Verify that every item on this checklist is addressed before exposing ACP to production traffic. Skipping any of these steps may leave your deployment vulnerable.

Item	Description
`ACP_API_KEY`	Set a strong API key for authentication on all endpoints.
HTTPS/TLS	Enable TLS termination at your reverse proxy or load balancer.
Database SSL	Configure PostgreSQL to require SSL connections.
LLM spending limits	Set spending caps on your OpenRouter, OpenAI, or Anthropic account.
Rate limiting	Configure rate limits at the edge and/or reverse proxy layer.
Logging and monitoring	Enable structured logging and set up alerts for error spikes.
Dependency updates	Establish a regular cadence for updating Python and npm dependencies.

Environment Variables

The following environment variables control security-related behavior:

Required

Required environment variables

OPENROUTER_API_KEY=sk-or-v1-xxx