A synthetic-media framework built as guidance, not a rulebook

The shot

In February 2023, the Partnership on AI published its Responsible Practices for Synthetic Media framework, the result of a process the organization says began in January 2022 and drew on more than 50 organizations: synthetic-media and technology companies including Adobe, Google, Microsoft, Meta, OpenAI and TikTok, broadcasters including the BBC and CBC, civil-society groups including WITNESS, Code for Africa and Thorn, and academic partners including Stanford HAI, per the Partnership on AI's own site.

What the documents show

The framework organizes its recommendations around three roles it names directly: builders of synthetic-media tools, creators who use them, and distributors who publish the results, with consent, disclosure and transparency as the recurring techniques across all three. It distinguishes direct disclosure, meaning viewer-facing labels, watermarks or disclaimers, from indirect disclosure, meaning cryptographic provenance data, metadata or steganographic signals embedded in a file. The organization is explicit that this is guidance, not enforcement: 'PAI will not be auditing or certifying organizations. This Framework includes suggested practices developed as guidance,' and PAI describes the document as functioning 'like a constitution, not a set of laws.' That is the organization's own account of a voluntary framework, and nothing in the document supports describing any named supporter as bound to it the way a contract or a regulation would bind them.

The workflow

For an organization that formally supports the framework, PAI's own site describes three concrete commitments: participating in a community of practice with other supporters, contributing a yearly case example or analysis showing how the framework applies to an actual product or technology, and joining one or two programmatic convenings a year where supporters discuss how the recommendations are holding up in practice. A production company or platform consulting the framework without formally joining it can still use the builder, creator and distributor categories as a checklist for where in a synthetic-media pipeline a disclosure decision needs to be made.

What the tool does not change

A framework organized around suggested practices does not, by the organization's own description, obligate any company to adopt a specific label, watermark or disclosure method, and it does not audit whether a signatory actually follows through. Whether a given production's disclosure choices meet the spirit of the framework remains a judgment made by the production itself, not something the document certifies.

• If this production used a synthetic-media tool, does the disclosure plan address builder, creator, and distributor responsibilities separately?
• Is a planned disclosure direct, indirect, or both, and does that match what the finished piece actually needs?
• Has any company invoked this framework in a way that claims more binding force than PAI's own description allows?

The framework is a documented, multistakeholder statement of recommended practice, not a rulebook, and PAI's own language is careful to keep that distinction visible.