WikiHow sues OpenAI over AI training content use

WikiHow has initiated legal action against AI firm OpenAI, accusing it of illicitly copying and utilizing WikiHow's copyrighted material for training its artificial intelligence models, including ChatGPT. The lawsuit alleges that OpenAI's development of ChatGPT involved the unauthorized use of WikiHow's extensive library of how-to instructions.

The core of WikiHow's complaint is that OpenAI exploited its content without a license or payment, thereby building a product that now directly supplies users with the substance of WikiHow's articles, diverting traffic and revenue from the original site. This practice, WikiHow argues, deprives them of essential page views that sustain their advertising income.

WikiHow contends that OpenAI has infringed on its copyrights through multiple avenues. These include large-scale copying for training data, incorporating content into retrieval-augmented generation (RAG) systems, and ultimately generating similar outputs for users. The company asserts that OpenAI's bots accessed WikiHow's pages hundreds of thousands of times within a short period and accuses the company of using obfuscated user agents and IP addresses to conceal its data collection activities.

While OpenAI has not publicly disclosed all datasets used, the lawsuit suggests that previous models like GPT-2 and GPT-3 incorporated material scraped from WikiHow. WikiHow claims that at least 11,211 of its articles were used to train GPT-4 and subsequent models. The suit seeks to compel OpenAI to cease using and storing its works and requests undisclosed monetary damages.

This lawsuit follows similar actions against AI companies. Earlier this year, The New York Times and Encyclopædia Britannica filed comparable lawsuits over alleged copyright infringement. However, in a separate case in April, the Delhi High Court denied an interim injunction sought by news agency ANI against OpenAI, finding that ANI's newer content was unlikely to have been part of GPT-4's training data.