STIM built an opt-in licence before writing a broader AI policy

The audio task

A songwriter affiliated with STIM, the Swedish performing rights society, wants a way to let an AI developer legally train on their catalogue and be paid for it, without waiting for national or EU legislation to settle the question first. STIM built a licence rather than only a policy statement, and its own site describes how that licence actually works.

What the documents show

STIM's 8 September 2025 announcement states STIM introduced what it calls the world's first collective AI licence for music, covering 'a selection of works by creators who have explicitly given their consent,' launched as a pilot with the startups Songfox and Sureel and built on master recordings from Cora Music rights holders. Interim CEO Lina Heyman is quoted describing it as ensuring 'anyone who wants to contribute their music to AI contexts is fairly compensated and has full transparency.' STIM's 'Stim and AI' position page, as retrieved on 16 September 2026, states that AI training on copyright-protected music 'requires a licence' and that royalty statements will show whether income came from AI training use or from generated output where a member's work had influence.

Rights status

Both documents describe an opt-in system: only works from songwriters and publishers who were individually asked and gave consent are included in this first version of the licence, and STIM states any expansion will happen 'in close dialogue with affiliated rights holders.' STIM does not state that AI training without a licence is unlawful everywhere; it states its own operating premise, that a licence is required to use STIM-represented music for AI training or output generation, and that the current licence's scope is deliberately limited while the model is tested.

What to check before you use it

This is an editorial checklist. A songwriter interested in participating should check whether their specific works are eligible for this pilot phase, since STIM states inclusion is currently limited to works individually consented into the Cora Music-linked catalogue. An AI developer evaluating STIM's licence should check what 'independent attribution technology' the licence requires for tracing AI-generated output back to source recordings, since STIM's announcement names this as a condition of use.

• Is my specific catalogue currently eligible for STIM's opt-in AI licence, or only for a future expansion?
• What attribution technology does the licence require, and how does it verify a generated output's source?
• How will STIM's royalty statements distinguish AI-training income from AI-output income in practice?

STIM's approach answers the licensing question for a limited, consenting group of members now, rather than waiting for a general rule, which is a different strategy from a federation-level advocacy statement and worth distinguishing when comparing collecting societies' AI responses.