Risks & safety
Copyright and AI
Also known as: AI and intellectual property
The unsettled legal questions about using copyrighted material to train AI, and about who owns what AI produces.
Copyright and AI is a big unresolved legal question with two sides. First: is it allowed to train AI on books, art and articles without permission? Creators and companies are in court over this. Second: who owns what AI makes, and can a purely AI-generated work even be copyrighted? The law is still catching up, and it varies by country, so anyone using generative AI commercially should stay aware of the risks.
AI raises two distinct copyright issues: 1. Inputs (training): models are trained on large amounts of copyrighted text and images, usually without explicit licences. Whether this is lawful (e.g. "fair use") is being litigated and differs across jurisdictions. 2. Outputs (ownership): who, if anyone, owns AI-generated content? In several jurisdictions, purely machine-generated works may not qualify for copyright, which needs human authorship.
Practical concerns
- Infringement risk: AI can sometimes reproduce protected content closely.
- Commercial use: businesses need to weigh licensing, indemnities offered by vendors, and disclosure.
It is a fast-moving area at the intersection of law, AI ethics and policy.
The copyright landscape for AI has two axes. On inputs, the core dispute is whether training on copyrighted works constitutes infringement or is permitted (e.g. US fair use, EU text-and-data-mining exceptions with opt-outs), the subject of active, unresolved litigation with divergent national approaches; memorisation/regurgitation of training data sharpens exposure. On outputs, authorship doctrines in several jurisdictions (notably the US Copyright Office's guidance) hold that purely AI-generated content lacks the human authorship required for protection, while human-directed works may be protectable in part.
Additional live issues: infringement when outputs are substantially similar to protected inputs, licensing and provenance of training data, style mimicry and right-of-publicity/deepfake concerns, and vendor indemnification. The situation is jurisdiction-specific and evolving through courts, regulators and legislation, so commercial generative-AI use warrants legal review rather than reliance on any single rule, and intersects with AI ethics, data governance and creator-compensation debates.