Comparative IP decisions
Proposals
A curated selection of the most significant decisions. Not an exhaustive database.
What 59 policy papers, resolutions and forums recommend on AI and IP, grouped by the ideas they share. These are proposals, not law.
Most widely backed · backed by 11 institutions in 5 regions
1Only human authors should get copyright in AI-generated output.
Policymakers 1/12 · Academia 8/34 · Professional bodies 2/6 · each group counts equally
Nuance: Machines aren't authors, but a thin three-year sui generis right could protect AI output.
Nuance: Sets the human bar very low: selecting and arranging prompts can be enough.
Differs: Proposes graded rights that grow with human involvement, rather than all or nothing.
Differs: Argues autonomous AI works fit copyright's structure more easily than assumed.
2AI developers should disclose what protected content they trained on.
6 institutions3Licensing, voluntary or collective, should be the main route to lawful AI training.
5 institutionsNuance: For webscraped content, prefers an exception with fair compensation over licensing.
Differs: Finds voluntary licensing impractical; proposes a mandatory blanket licence.
Differs: Prefers an output levy to training-based licensing.
4Inventions made with AI's help should stay patentable, with a human named as inventor.
4 institutionsNuance: Warns AI may already be the sole contributor, which the human-inventor rule doesn't cover.
Differs: Argues the AI itself should be recognised as the inventor.
5Creators should be paid when their work is used to build or run generative AI.
7 institutionsDiffers: Recommends no statutory licensing for now; leaves it to the market and fair use.
Differs: Training should generally be fair use, so no payment is owed for it.
6Today's opt-out from AI training doesn't work and should be fixed or replaced.
6 institutions7AI training on protected works should generally be allowed.
7 institutionsDiffers: Only with authorisation or under an applicable exception.
Nuance: Backs a data mining exception now, but would wait for the courts on LLM training.
Differs: Finds generative AI training is infringement that no current EU exception covers.
Differs: Argues every fair use factor weighs against training generative AI.
8Pay creators through a statutory scheme or levy rather than case-by-case licences.
4 institutionsNuance: Only for webscraped content; voluntary licensing for the rest.
On the agenda · forums with no position yet
Ranked by how many distinct institutions back each idea. Grouping is an editorial reading of the papers; open any paper to read it in full.