IS5542 Chap.11 AI Law: Copyright, Ownership and Licensing
AI Law: Copyright, Ownership and Licensing
Law as the environment
The legal guest lecture treats law as the production environment for AI: it governs consequences, not mechanics. Sources of law bind differently, from statutes and case law to regulations, contracts and non-binding guidance, and law lags technology, so courts apply existing rules to new facts after harm occurs.
Ethics principles are not compliance, and presenting them as such is ethics washing.
Four kinds of intellectual property
Copyright protects original expression automatically; patents protect registered inventions but not algorithms as such; trade secrets protect confidential information such as model weights; trademarks protect brand identifiers now threatened by deepfakes.
Copyright protects expression, not ideas, which is why a model's learned patterns are ideas while its training texts are expression.
The training-data dispute
Rights holders argue that training copies works and that outputs replace the original market; developers argue that training transforms and that fair use or mining exceptions apply.
The United States weighs four fair-use factors; the EU allows commercial mining unless rights holders opt out; the UK allows only non-commercial mining; the Mainland has no broad exception.
Who owns the output
Protection depends on human input and on jurisdiction: US law requires human authorship, a Beijing court found copyright where a user's prompting was creative, and UK law protects computer-generated works.
Licences and provenance labelling fill the gaps in practice.
What this chapter covers
- 01
Sources of law and how they bind
- 02
The lag between technology and law
- 03
Ethics principles versus regulation
- 04
Four pillars of intellectual property
- 05
Idea and expression
- 06
Arguments over training data
- 07
Fair use, mining exceptions and opt-outs
- 08
Authorship across jurisdictions
- 09
Code licences, weights and provenance
Worked example · free
Assess an opt-out for a news publisher
- 3Explain the commercial mining exception and its opt-out.
- 2State the lawful-access condition.
- 3Recommend further steps.
Key terms
- Idea-expression Dichotomy
- The principle that copyright covers how an idea is expressed, leaving the idea itself, facts and functional processes free.
- Fair Use
- A US defence to copyright infringement decided by balancing the purpose, the nature of the work, the amount taken and the market effect.
- Text and Data Mining
- Automated analysis of large bodies of content, permitted in the EU for research and, unless opted out, for commercial purposes.
- Ethics Washing
- Presenting voluntary AI ethics principles as though they amounted to legal compliance.
- Copyleft Licence
- An open-source licence, such as the GPL, requiring derivative works to be released under the same terms.
- Content Credentials
- Cryptographic provenance metadata recording how a piece of content was created, as in the C2PA standard.
- Trade Secret
- Confidential business information with commercial value protected by reasonable secrecy measures, which covers model weights.
AI Law: Copyright, Ownership and Licensing FAQ
Can I own the copyright in an image I generated with AI?
It depends on where and how. The US requires human authorship and has refused works credited to an AI system, while a Beijing court recognised copyright where the user's choice of prompts and settings was original, and UK law protects computer-generated works for fifty years. Documenting your creative input strengthens any claim.
Does training a model on copyrighted text infringe copyright?
Courts have not yet given a final answer. Rights holders say training copies works and that outputs substitute for them; developers rely on fair use in the US or mining exceptions elsewhere. Evidence that a model reproduces training text almost word for word weakens the argument that training is purely transformative.
Why does the lecture say ethics principles are not a legal shield?
Because voluntary principles are enforced by reputation, not by courts, and carry no weight in litigation. Binding rules such as the EU AI Act's prohibitions are enforced by regulators with fines of up to 35 million euros or 7% of global turnover, so a charter cannot replace compliance.
Exam move
Build a four-column table for the US, the EU, the UK and the Mainland with rows for training-data rules, authorship of output and the main case or instrument. Then answer one short question per row aloud, naming the jurisdiction first each time.
Working through AI Law: Copyright, Ownership and Licensing in IS5542? Sia is AskSia’s AI Information Technology tutor — ask any IS5542 AI Law: Copyright, Ownership and Licensing question and get a clear, step-by-step explanation grounded in how IS5542 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.
