CISAC's 2023 AI principles ask for consent, not just credit

The audio task

A songwriter registering a work with a national authors' society wants to know one practical thing: if a generative AI system trains on that catalogue, does anyone have to ask first, and does the songwriter see any money from it. CISAC, the International Confederation of Societies of Authors and Composers, groups 227 member societies representing more than five million creators, and its own site sets out what it is asking governments to require before that question gets answered by default.

What the documents show

CISAC's Artificial Intelligence resource page states that its advocacy rests on three principles: authorisation, meaning creators must have the right to licence their works; remuneration, meaning creators must be paid; and transparency, meaning AI providers must disclose what copyrighted works trained their systems. The same page records that in July 2023 CISAC set out seven principles for policymakers, including that creators' and performers' rights must be upheld when exploited by AI systems, that licensing should be enabled, and that legal responsibility should attach to AI operators. A separate policy page describes a dedicated CISAC working group on AI and a 2020 submission to the World Intellectual Property Organisation, and states CISAC continues monitoring legislative initiatives in the United States and Europe. Neither page claims any government has adopted these seven principles as law.

Rights status

What CISAC has published is an advocacy position from a federation of collecting societies, not a licence, a ruling or an enacted rule. The pages state what CISAC is asking legislators and AI operators to do; they do not state that any AI developer is currently bound to seek consent, pay remuneration, or disclose training data as a matter of law in any specific country. Whether a given national member society, such as one representing a specific territory, has itself achieved any of these outcomes in domestic law is a separate, jurisdiction-by-jurisdiction question that this statement does not answer.

What to check before you use it

This is an editorial checklist, not legal advice. Before treating CISAC's principles as a rights guarantee, a producer or catalogue manager should check whether their own national authors' society has published a matching position, since CISAC represents many independent societies rather than issuing binding instructions to them. It is also worth checking whether a specific AI vendor's own licence or training-data disclosure, if any, actually reflects authorisation, remuneration and transparency, rather than assuming CISAC's asks have already been met.

• Has my national collecting society published its own AI position, and does it match CISAC's three principles?
• Does the AI tool I am evaluating disclose, in its own documentation, whether it licensed its training catalogue?
• Is there an enacted law in my jurisdiction that mirrors CISAC's asks, or is this still advocacy?

CISAC's own pages are useful for understanding where a large bloc of authors' societies wants AI policy to go, but they describe a destination, not a current legal state, and a rights holder should treat the distinction as load-bearing.