AI and copyright: key issues for startups

Generative AI raises new copyright questions as it produces text, images, music and code. The law is contested and evolving in many countries. Startups should understand the risks and take reasonable precautions.

Training data

The copyright status of data used to train models is subject to lawsuits and regulatory debate worldwide. If you train or fine-tune your own model:

• Document the source and licence of your data.
• Use customer data only as contracts allow.
• Prefer openly licensed or self-generated data; see AI data strategy.

Are AI outputs protected?

Copyright generally protects human creative contribution. Protection for content generated entirely by AI is uncertain in many countries; it becomes more likely where human selection, arrangement and creative input are significant. Turkish law also ties authorship to natural persons, so documenting human contribution in important content helps; see our IP guide.

Infringement risk

• Models can produce content very similar to training data.
• Prompts involving well-known characters, brands, artists' styles or real people's likeness are risky.
• Generated code may resemble licensed open-source code; see AI in software development

Provider terms

Check model providers' terms for:

• Who owns the outputs?
• Does the provider offer indemnity against copyright claims, and on what terms?
• Are your inputs used to train the model?

Practical safeguards

1. Make human creativity and editing part of the process for important content.
2. Avoid prompts aimed at imitating other works, brands or people.
3. Run similarity checks before publishing; see AI content creation.
4. Clarify output rights and responsibilities in your product's terms; see our terms of use guide.
5. Follow developments and get legal advice when needed.

Conclusion

AI and copyright are changing fast. Documented processes and careful use protect startups. This guide is general information; consult a lawyer on specific cases.