Origin Story
This started with a simple question.
When AI replaces the UI, who enforces the disclaimers?
For decades, data providers have relied on a simple assumption: a human developer will build a frontend, read the documentation, and manually place the required disclosures, formatting rules, and attribution notices next to the data. The compliance layer lived in the UI because a person built the UI.
That assumption is breaking. AI agents are increasingly the ones presenting data to end users. They generate responses dynamically. There is no hardcoded frontend. No developer reading the terms of service. No human in the loop deciding where to put the "past performance" disclaimer.
The question seemed obvious. The answer was: nobody. There was no protocol, no standard, no mechanism for data to carry its own presentation rules into an AI-mediated world.
AIPC exists to fill that gap. It is a contract envelope that wraps data with machine-executable rules -- disclosures, formatting, attribution, freshness, tone constraints -- so that any AI system, from any vendor, can present that data correctly. Or stay silent.
The data should carry its own rules. Always.
Design Principles
Five ideas that shape every decision.
These principles are not aspirational. They are constraints. Every feature in the spec must satisfy all five, or it does not ship.
Data carries its own contract
Every API response includes the rules for how that data must be presented. No external documentation to read, no terms of service to parse, no assumptions to make. The contract travels with the data, inseparable from it.
Progressive complexity
A simple dataset can ship with a simple contract: just attribution and a disclosure. A complex financial instrument can ship with conditional rules, jurisdiction gates, and field-level formatting. You only use what you need.
AI-runtime agnostic
AIPC works with any AI system: LLMs, retrieval-augmented pipelines, tool-calling agents, voice assistants. The contract is pure JSON. If your system can read JSON, it can enforce AIPC contracts.
Human-readable and machine-parseable
A compliance officer should be able to read a contract and understand what it requires. A runtime should be able to parse and enforce it programmatically. AIPC achieves both without sacrificing either.
Fail-safe by default
If an AI runtime cannot comply with a contract, the default behavior is suppression: the data is not presented at all. This is the trust mechanism that makes compliance teams comfortable. Silence is always safer than a wrong answer.
People
Team & Contributors
AIPC is early. Right now it is one person and a conviction. That will change.
Geoffrey
Creator & Spec Author
Conceived the protocol after observing the gap between regulated data APIs and AI-generated interfaces. Building the spec, the reference runtime, and the community from scratch.
Future contributors welcome
AIPC needs people who understand regulated data, AI runtimes, compliance frameworks, and developer tooling. If that sounds like you, there is a seat at the table.
Learn how to contributeLicense
Open by design.
AIPC is an open protocol. The spec and the code are licensed separately to maximize adoption.
CC BY 4.0
The AIPC specification document is licensed under Creative Commons Attribution 4.0 International. You are free to share, adapt, and build upon the spec for any purpose, including commercial use, as long as attribution is given.
View licenseMIT
The reference runtime, validators, SDK, and all code artifacts are released under the MIT License. Use them in any project, commercial or otherwise, with no restrictions beyond preserving the copyright notice.
View licenseFAQ
Common questions.
Straightforward answers. If your question is not here, open a GitHub Discussion.
Still have questions?
Open a Discussion on GitHub