Skip to content

B.C. author claims AI company used his work without consent

Class-action suit calls use of copyrighted material for computer learning without consent "high-handed" and "arrogant."
aicomputer
AI — or artificial intelligence — uses vast amounts of data for machine learning.

A B.C. author is proposing a Canadian class-action lawsuit against a San Francisco-based artificial intelligence company alleging it used his and other people’s copyrighted work to train its computer system.

In a May 23 B.C. Supreme Court notice of civil claim, James Bernard MacKinnon alleges that Anthropic PBC set out to build an artificial intelligence large language model (LLM) — a technology trained on large amounts of text data to learn nuances of language.

Such models can be used for tasks like creative writing, translation or question answering.

“Anthropic has benefited from its use of the copyrighted work without licence or permission as that work allowed it to improve the potentially extremely lucrative Anthropic LLMs,” the claim said.

The lawsuit said the alleged conduct was “high-handed, arrogant, and displayed a reckless disregard for the rights of the class members.”

The claim asserts Anthropic chose to train its model on a library of books with the belief it could create an advantage over competitors.

But, the claim asserts, the company used a dataset it found online that it knew contained unlicenced and copyrighted material.

“It made no effort to pay the owners of this copyright to obtain or use their works,” the claim said.

MacKinnon said his works are among those used without consent, licence or permission. So, he filed the claim for copyright and unjust enrichment against the company as a possible class action on behalf other others who may be in the same situation.

The claim said Anthropic was founded by former employees of Open AI Inc. in January 2021 and has since received investment from Amazon.com Inc. and Google and/or companies affiliated with those entities.

The claim said Anthropic fed data into its LLM to train it.

Then, in March 2023, the claim said, Anthropic released an LLM product called Claude.

The claim said the first version of Claude was trained using a dataset called Books3 that Anthropic downloaded from the internet without paying a licensing fee.

MacKinnon said in the claim that a book he authored and another he co-authored were in that dataset.

He said those works and works of others is copyrighted work belonging to him and other Canadian authors.

Further, the claim said, Books3 did not license the copyrighted material in any way.

The claim alleges Anthropic took steps to conceal its use of copyrighted material in such a way that the LLM itself did not know if was built off copyrighted material.

And, the claim said, Anthropic has grown to become a US$60-million company.

The suit seeks damages, part of profits received for use of copyrighted material, disgorgement of profits, punitive damages and costs.

Lodestar Media has reached out to Anthropic for comment but received no response.

None of the allegations have been proven in court.