KazakhstanLaw on Artificial Intelligence, machine-readable opt-out for training on published works
Law No. 230-VIII (), Art. 23(5)
In force since .
A text and data mining (TDM) opt-out rule binding public and private bodies.
What it requires
- Before using works, published news articles included, to train an AI model, check whether the author or rightsholder has expressed a prohibition in machine-readable form, and do not train on the work if they have (Article 23(5)).
- Do not rely on copyright's free-use exceptions for educational or scientific purposes to justify training: training does not fall within them (Article 23(3)).
- Training a model does not itself amount to exercising the author's reproduction, distribution, adaptation, public display, performance or broadcast rights (Article 23(4)), but keep a separate licence or basis for your underlying reproduction of the articles themselves; the machine-readable check does not replace it.
These lines are LexLint's reading of the law's text.
In the frameworks
- NIST AI RMF
- MAP 4.1Approaches for mapping AI technology and legal risks of its components – including the use of third-party data or software – are in place, followed, and documented, as are risks of infringement of a third party’s intellectual property or other rights.
Our reading of where this law's duties sit in each framework. It never says that running a control meets the law. How we read a framework against the law
What this law does
Article 23(5) of the Law on Artificial Intelligence permits using works, published news articles included, to train an AI model only where the author or rightsholder has not expressed a prohibition in machine-readable form.
The same Article separately states, at Article 23(3), that training does not fall within copyright law's free-use exceptions for educational or scientific purposes, and at Article 23(4) that training a model does not itself amount to exercising the author's reproduction, distribution, adaptation, public display, performance or broadcast rights.
Read together, an aggregator that trains a model on the news content it aggregates needs the absence of a machine-readable rightsholder objection under Article 23(5), separately from whatever licence its underlying reproduction of the articles themselves rests on.
When LexLint raises it
When your app profile says your app trains models, reuses other publishers' content or crawls the web.
Back to the example · Lint your app
Page updated .