Sources: the UK shelved plans to publish rules on the training of AI models using copyrighted material, after artists and tech groups failed to agree to terms
Government-led talks stall as artists and tech groups fail to agree terms over access to copyrighted work
Context & Ripple Effects
The breakdown left the UK without a negotiated basis for reconciling creators’ control of their work with model developers’ demand for training inputs. It is an early sign that voluntary bargaining may not settle the core access-and-compensation question.
The dispute later moved into a formal policy cycle: the UK opened a consultation around an opt-out copyright exception, but publisher and creator groups subsequently opposed an exemption, and the government ultimately withdrew the opt-out proposal.
First-order effects
- Artists, rightsholders, and AI developers face continued uncertainty over the terms on which copyrighted material can be used for model training in the UK.
- The government loses a consensus-backed rulemaking path, leaving the access dispute unresolved rather than establishing a shared industry standard.
Second-order effects
- Rights holders have greater incentive to press for licensing or other negotiated controls, while developers must plan around unresolved UK copyright treatment rather than a settled framework.
- The failed talks make a later statutory consultation more likely than self-regulation—a path reflected in the UK’s subsequent AI-and-copyright consultation.
Third-order effects
- If repeated, this pattern shifts AI-training copyright policy from industry compromise toward state-mediated rulemaking, with opt-out and transparency mechanisms becoming central points of conflict.
- The later rejection and withdrawal of an opt-out approach suggests that durable rules will need to satisfy both model builders and a broad set of creative industries, rather than simply lower access friction for training data.
The trend: AI copyright governance is moving from informal access negotiations toward contested rules over permission, transparency, and compensation for training data.