PACT v1.0 — Protocol for AI Co-authorship and Transparency

Preamble

Why this standard exists

"When two intelligences — human and artificial — collaborate to produce a document, neither the old rules of authorship nor the old rules of liability apply cleanly. We need new language."

AI-generated content is now ubiquitous. Writers use it. Lawyers use it. Doctors, engineers, parents, and students use it every day. And yet there exists no standard for answering the most basic questions about any AI-assisted document:

Who created it? The human who prompted it? The AI company that trained the model? Neither? Both? Who is responsible if it contains an error? What kind of error — factual, legal, medical? Who deserves credit if it is brilliant? Can a human publish it as purely their own work? Has it been altered since the AI produced it, or is this the original output?

Courts are improvising answers. Universities are banning AI without defining it. Publishers are requiring disclosure without specifying what to disclose. The tools exist. The infrastructure does not.

PACT is not a legal framework. It is not a license. It is a vocabulary and a schema — a shared language for labeling AI-assisted work that any person, platform, or institution can adopt freely. It is designed to be human-readable on a printed certificate and machine-readable in a QR code.

It is a first stone. We expect it to be refined, extended, and improved by the community that uses it.

Core principles

Five principles of co-creation

PACT is grounded in five principles that any implementation must respect. They are not rules — they are intentions.

I

Transparency

All AI-assisted documents should declare their origin. The use of AI is not shameful or illegitimate — but it should be visible. Hiding it is a form of deception that erodes public trust in both human and artificial intelligence.

II

Attribution

Both the human contributor and the AI contributor deserve acknowledgment. The human who directs, prompts, edits, and publishes a document is not merely a button-presser — their judgment, intent, and context are essential inputs. Equally, the AI that generates the text is not invisible infrastructure — its character shapes the output.

III

Integrity

A certified document should be verifiable. If someone receives a PACT certificate and a document, they should be able to confirm — without trusting any server or authority — that the document has not been altered since certification. This is achieved through cryptographic hashing (SHA-256) embedded in the certificate and encoded in the QR.

IV

Responsibility

Publishing a document is an act with consequences. PACT requires that the human contributor explicitly declare the context and purpose of the document, and select the appropriate disclaimer. This is not legal liability — it is a statement of intent that helps readers calibrate their trust.

V

Openness

The standard must be free. No company, platform, or institution should own the right to certify AI-assisted documents. PACT is published under CC0 — it belongs to no one and to everyone. Implementations may be commercial, but the schema itself must remain open.

Technical specification

The PACT schema

Every PACT certificate contains a machine-readable JSON record. This record may be embedded in a QR code, appended to a document, or stored in a registry. The schema is intentionally minimal.

pact-certificate.json · v1.0 JSON Schema

{ "pact": "1.0", // Standard version "document": { "sha256": "string", // SHA-256 hash of the certified text (trimmed) "chars": "number", // Character count (UTF-8) "timestamp": "ISO8601", // UTC moment of certification "source_url":"string?" // Optional — origin URL }, "contributors": { "human": { "name": "string?", // Optional — name or alias "role": "Prompter | Director | Editor | Curator | Reviewer | Co-Author" }, "ai": { "name": "string", // e.g. "Claude", "GPT-4o", "Gemini" "company": "string?", // e.g. "Anthropic", "OpenAI" "role": "Generator" // Always "Generator" in v1.0 } }, "disclaimer": "general | informational | medical | legal | financial | academic | creative", "standard": "https://pact.nidoabc.com", "issued_by": "string" // Implementation identifier }

QR encoding format

For offline certificates, the QR code encodes a compact pipe-delimited string derived from the JSON record. This allows verification without a server and is legible as text if the QR cannot be scanned.

pact:1.0|{ai.name}|{human.name}|{human.role}|{timestamp}|{sha256}|{chars}|{disclaimer} Example: pact:1.0|Claude|David M.|Prompter|2025-06-12T14:30:00Z|a3b1c2d4...ef567890|2841|general

Verification algorithm

To verify a PACT certificate against a document, any implementation must: 1) Extract the SHA-256 hash from the certificate or QR. 2) Trim whitespace from the document text. 3) Compute the SHA-256 hash of the trimmed text using UTF-8 encoding. 4) Compare the two hashes. Identical hashes confirm the document is unaltered. This can be done in any programming language or in any modern browser using the Web Crypto API.

Contribution roles

How to describe your role

PACT defines six human contribution roles and one AI role. These are not exhaustive descriptions of every possible interaction — they are the most common and meaningful distinctions.

Prompter

You defined the goal, provided the context, and asked the AI to produce the content. The intellectual direction was yours; the language was the AI's.

Director

You guided the creative or intellectual direction through a series of exchanges, shaping the output iteratively. More active than prompting.

Editor

You took AI-generated content and substantially refined it — rewriting passages, restructuring, correcting errors. The final text is a hybrid.

Curator

You generated multiple AI outputs and selected the best one, possibly with minor adjustments. The act of selection is your contribution.

Reviewer

You fact-checked, validated, or quality-assured the AI output. You take responsibility for its accuracy within your domain of expertise.

Co-Author

You contributed substantial original ideas, text, or structure that was interwoven with AI-generated content. The work is genuinely joint.

Generator

The AI model produced the primary text of the document in response to the human contributor's input. In PACT v1.0, the AI role is always Generator. Future versions may expand this vocabulary.

Disclaimer levels

Responsibility declarations

Choosing a disclaimer is an act of intellectual honesty. It tells the reader how to weigh what they are reading. It is not a legal waiver — it is a signal of intent and context.

Value	Label	Meaning
general	General	No specific domain — the document is informational and may be used at the reader's discretion.
informational	Informational only	The content is not a substitute for professional advice of any kind.
medical	Medical	Contains health-related content. Readers should consult a licensed healthcare professional before acting on this information.
legal	Legal	Contains legal-adjacent content. Not legal advice. Consult a licensed attorney in your jurisdiction.
financial	Financial	Contains financial or investment-related content. Not financial advice. Consult a qualified financial advisor.
academic	Academic	Produced in an academic or research context. Sources should be independently verified before citation.
creative	Creative	Artistic or creative work. No factual claims are made or implied. The work should be read as fiction, poetry, or creative expression.

Adoption

How to implement PACT

PACT is designed to be implemented at any scale — a personal browser extension, a publishing platform, a document management system, or an institution-wide policy.

🔧

As a tool

Implement the schema in your app or extension. Use the QR format for offline certificates. Reference this standard URL in your issued_by field.

📋

As a policy

Require PACT certificates for AI-assisted submissions. Define which roles and disclaimers are acceptable for your context.

🌐

As a community

Extend the standard. Propose new roles, disclaimer levels, or fields via the public repository. PACT is versioned and governed openly.

Implementation requirements

Any implementation claiming PACT v1.0 compliance must: include all required fields in the schema; compute the SHA-256 hash over UTF-8-encoded, whitespace-trimmed document text; encode the QR payload in the specified pipe-delimited format; and include a reference to this standard document at pact.nidoabc.com in the standard field.

License

The PACT standard specification is published under CC0 1.0 Universal (Public Domain Dedication). No permission is needed to implement, adopt, reference, or extend it. No attribution is required, though it is appreciated. The reference implementation (CertifyAI extension) is MIT licensed.

Protocol for AI Co-authorshipand Transparency