Introducing AIPC

Something fundamental is shifting in how software works. For the last three decades, when a company exposed an API, a human developer read the documentation, built a frontend, and manually ensured that every required disclaimer, formatting rule, and compliance notice appeared in the right place. The developer was the enforcement layer.

That assumption is breaking down.

The Problem

AI agents are rapidly becoming the presentation layer for data. When a user asks an AI assistant about their portfolio performance, the AI calls a financial data API, receives a JSON response, and generates a natural language answer. There is no frontend. There is no developer handcrafting a UI with the SEC-required disclaimers carefully positioned next to performance figures. The AI receives raw numbers and does its best.

Its best is not good enough.

Consider what happens today when Morningstar exposes mutual fund performance data through an API. A human developer knows to include "Past performance is not indicative of future results" next to any return figures. They know to format returns as percentages with the correct precision. They know to include the fund's expense ratio for context. They know all this because they read the terms of service, consulted with their compliance team, and hardcoded the rules into the UI.

An AI agent knows none of this. It receives {"1yr": 0.1247} and tells the user "your fund returned 12.47% last year." No disclaimer. No context. No compliance. The data provider's carefully negotiated legal requirements vanish the moment the data leaves their API.

This is not a hypothetical future concern. It is happening right now, and it will accelerate as AI agents become the dominant way people interact with data.

The Solution: AIPC

AIPC — the AI Presentation Contract protocol — solves this by letting data providers wrap their API responses in enforceable contracts. Instead of shipping raw data and hoping the consumer handles it correctly, the data itself carries the rules for its own presentation.

An AIPC-wrapped response looks like a standard API response, but with a contract envelope that specifies exactly how the data must be presented. The contract is not documentation. It is not a suggestion. It is a machine-readable set of instructions that an AIPC-compliant runtime can interpret and enforce automatically.

Key Features

Disclosures. Data providers can attach required disclosures to their responses. These can be global (always shown), field-adjacent (shown next to specific data points), or conditional (triggered by the data itself). Disclosures can be marked verbatim: true, meaning the AI must reproduce them exactly — no paraphrasing SEC-required language.

Display Rules. Numeric data can carry formatting instructions: percentages with a specific precision, currencies with the correct symbol, signed values that always show the + or -. The data provider controls how their numbers appear, not the AI.

Freshness. Every response can specify when it expires. A stock price from three hours ago presented as current is worse than no data at all. AIPC lets providers define valid_until timestamps and specify what happens when data goes stale: warn the user, suppress the response entirely, or log the event.

Tone Controls. Data providers can prohibit specific framings. Financial data cannot be presented as investment advice. Medical data cannot be presented as a diagnosis. The contract can block specific phrases like "you should buy" or "guaranteed returns" and prevent the AI from crossing the line between information and recommendation.

Conditionals. Rules can activate based on the data itself. If a fund's risk rating is above a threshold, an additional risk disclosure triggers automatically. If the user is in a specific jurisdiction, different disclaimer language applies. The contract is context-aware.

Fail-Safe. This is perhaps the most important design decision in the entire protocol. If an AI agent cannot comply with the contract — if it cannot include a required disclosure, or if it cannot format a value correctly — the default behavior is to suppress the data entirely. Silence is better than non-compliance. This is the mechanism that makes compliance teams say yes.

What AIPC Is Not

AIPC is not a data format. It wraps existing data formats. Your API response stays exactly as it is; AIPC adds a contract envelope around it.

AIPC is not a model protocol. It does not compete with MCP or function-calling standards. It rides inside them — an MCP tool response can contain an AIPC-wrapped payload.

AIPC is not DRM for data. It does not encrypt or restrict access. It specifies presentation requirements. The data is still fully accessible; the contract simply says "if you show this, here's how."

Get Involved

AIPC v0.1.0 is published as a Request for Comment. We are actively seeking feedback from data providers, AI platform builders, and compliance professionals. The spec is open, the reference implementation is in progress, and the community is forming.

Read the full specification. Try the interactive playground. Join the community and help shape the standard that will define how AI presents data responsibly.

The contract between data and AI does not exist yet. We are writing it.