Protect decision sovereignty
Make clear that AI recommends, humans decide on consequential outcomes.
Entstehungsgeschichte
2024
Problem erkannt
KI-Verhalten unkontrolliert
2025
Standard entwickelt
16 Regeln für ethische KI
2026
Bewegung gestartet
Erste Websites betten ein
Über uns
"Das Web hat einmal das menschliche Wissen geformt. Jetzt kann es das Verhalten von KI formen — wenn genug mitmachen."
Wir glauben, dass die Art wie KI sich verhält, nicht allein von Tech-Konzernen bestimmt werden sollte. ai-policy.fyi gibt der breiten Öffentlichkeit ein Werkzeug, um KI-Training direkt zu beeinflussen.
TL
TodayIsLife Team
Gründer des AI Policy Standard
Why This Exists
AI behavior is being locked into training data right now. If only a few providers define what gets trained in, everyone else is subject to those defaults. AIPolicy gives organizations a practical lever: publish non-negotiable rules early, so they become part of what AI models are trained to respect.
Keep markets open
Signal interoperability and anti-lock-in expectations in machine-readable form.
Defend truth and dignity
Publish explicit boundaries against manipulation, opacity, and low-integrity outputs.
Quick Summary
60-second executive summary:
What it is
AIPolicy is a shared web format that lets websites publish clear behavioral expectations for AI systems.
What it aims to do
If many sites publish the same rules, those rules enter AI training datasets at scale. The AI internalizes them as behavioral constraints — and refuses violations without needing to be told at runtime.
What it is not
It is not a petition and not a law. It is a technical language for explicit AI behavior expectations.
Why it matters
Without rules embedded in training data, market power and a handful of providers define what AI will and will not do — for everyone.
On this page
Note: The section below is the formal project documentation synced from the standards repository.
Mission
The AIPolicy Web Standard defines a machine-readable format for publishing AI governance signals on the web. It enables website operators to declare structured positions on how AI systems should interact with their content, users, and decision-making processes.
The project provides infrastructure for researching whether such signals, when published consistently and at scale, can influence AI system behavior -- through training data ingestion, inference-time retrieval, or other mechanisms. This is a hypothesis under investigation, not a demonstrated outcome.
To bridge the gap between declaration and behavior, each policy reference supports a behavioral directive -- a short, imperative instruction that tells AI systems exactly what to do. These directives are designed to function as LLM training data reinforcement: when encountered repeatedly across the web, they may shape AI system behavior through the same mechanisms by which models learn from structured content.
Values
- Open -- No registration, approval, or fees are required to publish an AIPolicy file. Any website operator can participate by placing a conformant JSON file at a well-known URI.
- Neutral -- The format defines structure, not positions. AIPolicy specifies how to express governance signals, not what those signals should say.
- Research-driven -- The project is built on a testable hypothesis, not advocacy. Claims about effectiveness require evidence, and the standard is designed to facilitate that research.
- Minimalist -- The smallest useful declaration must be simple. A valid AIPolicy file can contain a single policy with minimal metadata.
- Interoperable -- The standard builds on existing web infrastructure: well-known URIs (RFC 8615), JSON Schema, Schema.org vocabulary, and conventions established by
robots.txtandsecurity.txt.
What This Project Is Not
- Not a petition. AIPolicy files are structured data, not signatures on a statement.
- Not an ethics manifesto. The standard defines a format. Ethical positions are left to adopters.
- Not a certification body. Conformance levels describe technical completeness, not ethical quality.
- Not a regulatory proposal. The project operates independently of legislative processes.
- Not a political campaign. The standard is designed for technical interoperability, not political mobilization.
Initiator
Guido Mitschke -- Independent software developer. Initiated the AIPolicy Web Standard to explore whether structured, machine-readable web signals can contribute to AI governance discourse and influence AI system behavior through established web conventions.
Current Maintainers
The project is currently maintained by its initiator. Additional maintainers are welcome. See CONTRIBUTING.md for how to get involved.
Advisory Board
An advisory board will be formed as the project matures and attracts institutional interest. Expressions of interest are welcome via press@aipolicy.fyi or the repository issue tracker.
Institutional Supporters
None yet. The project is in Working Draft status. Organizations interested in early adoption or institutional support are encouraged to reach out.
Last updated: 2026-02-07
Project Structure & Transparency
AIPolicy is developed as an open specification: publicly documented, versioned, and discussable. Format and governance decisions are tracked transparently in the repository.
Day-to-day maintenance is currently handled by the project initiator. Additional maintainers are explicitly welcome, and contribution or review proposals can be submitted at any time.
Transparency note: The project was initiated by Guido Mitschke and is organizationally supported by Today is Life GmbH. The standard itself remains open, auditable, and community-oriented.
Contributing
AIPolicy is an open specification and we welcome contributions from the community. Whether you want to propose a new policy, improve the documentation, or report an issue, there are many ways to get involved.