Know what your AI can do. Control what it's allowed to do. Prove what it actually did.

Your team is using Claude, ChatGPT, Gemini, or Grok. Without governance, one compromised connection, one tired engineer, one angry customer can break compliance or lose data. Guardrails makes it explicit: what this AI can access, what it's allowed to do, and how we prove it followed the rules. 45-minute workflow. Machine-readable policy. Works for startups and enterprises. Free. Open source.

View on GitHub See the authority ladder

Status: Phase 1 — live Launch: September 2026 Schema v1.0 — stable ontology License: MIT


GitHub stars0

GitHub forks0

Open issues0

Live sinceSept 10, 2026

Any AI, anywhere it shows up

Works with what you already use.

Guardrails doesn't care which AI you use — it cares what it can access and what it's allowed to do. That includes AI you didn't realize you had.

Where it shows upExamplesWhat it's typically used for
Direct AI toolsClaude, ChatGPT, Gemini, Copilot, Grok, PerplexityDrafting, analysis, code, research, customer support
Embedded in tools you already ownSalesforce Einstein, Microsoft 365 Copilot, Zendesk AI, GitHub CopilotCRM predictions, meeting summaries, support answers, code suggestions
Internally built systemsCustom agents, RAG systems, API calls to Claude or OpenAIDocument review, incident response, forecasting bots
The problem

AI authority is expanding faster than governance.

Six ways this is happening right now, inside companies that never decided to take these risks.

3:00 AM

The DBA

Production database down. Panicked. Asks Claude to fix it. Claude has prod credentials. No guardrails. Database corrupted. Nobody knows: Claude did it? DBA did it? No audit trail.

10:00 AM

The recruiter

Uploads 50 applicant résumés to ChatGPT to screen candidates. Personal ChatGPT account. No governance. Applicant data now on OpenAI servers. GDPR violation.

Quarter-end

The finance analyst

Pastes unreleased financial projections into Claude to "sense check" the forecast. Quarter-end projections are material non-public information. Insider trading exposure. Regulator liability.

Support queue

The support agent

Customer is furious. AI suggests an exception to company policy. Support agent accepts it. Now the company honored a policy exception. Customer expects it. Company liable.

Sev-1

The autonomous ops agent

Deployed to handle Sev-1 incidents. Has prod access. Incident occurs. Agent attempts action outside its intended scope. Security controls didn't exist. System compromised.

Breach

The compromised AI system

Critical infrastructure AI should never modify security settings. Attacker compromises the AI's infrastructure. AI gets weaponized. Operator can't tell if it's been exploited. No audit trail.

The AI actor's information access or authority exceeds the controls appropriate to the risk.
— the problem, in one line
AI authority must be explicit, bounded, observable, and proportionate to risk.

Know

What AI systems does your company use? What can they access? What are they allowed to do?

Control

How do we actually enforce those boundaries? In the system itself, not just in policy?

Prove

When something happens, can we audit exactly what occurred, who authorized it, and whether controls worked?

The authority ladder

Seven graduated levels of AI authority.

Same ladder works for code changes, financial decisions, hiring recommendations, manufacturing operations, critical infrastructure, and everything else.

LevelAuthorityMeaning
0ObserveAccess approved information; no derived action
1AnalyzeInterpret, classify, summarize, or diagnose
2RecommendPropose a decision or course of action
3PrepareCreate an artifact/change that cannot take effect without another actor
4Execute BoundedPerform predefined, reversible actions within explicit constraints
5Execute GatedPerform higher-impact actions only when required conditions/approvals are satisfied
6ProhibitedAction is unavailable to the AI actor regardless of instruction
What you get in 45 minutes

Four outputs from one workflow.

01

AI System & Use Case Register

Source of truth. Every AI system your company uses, every use case, classified.

02

AI Governance Policy

Human-readable baseline policy document, one section per use case.

03

Control Gap Assessment

What's missing? Prioritized by risk, with fix-by timelines.

04

Evidence & Validation Plan

How we prove this works: tests, owners, and review cadence.

Who this is for

Built for companies without a governance team.

You don't need a CTO or a security title to use this. You need to be worried about how AI is being used at your company and not know where to start.

Primary

  • Business owners and leaders at companies with no dedicated security, IT, or compliance person
  • CTOs, VPs of Engineering, and security/compliance leaders bringing structure to AI use
  • Any company — 5 people or 5,000 — deploying Claude, ChatGPT, Gemini, or Grok without governance

Secondary

  • Enterprises with GRC teams, using Guardrails as an operational/control translation layer
  • Regulated industries: healthcare, finance, legal, critical infrastructure

For developers

Guardrails is the governance framework beneath NIST, ISO, and OWASP. Not replacing them. Implementing them. We turn abstract AI governance requirements into company-specific controls, enforcement rules, and audit evidence.

View on GitHub