Key Insights:

• AI companies hit a “data frontier,” forcing them to explore new, costly methods for acquiring training data.
• Anthropic and OpenAI face lawsuits over alleged unauthorized use of copyrighted content in their AI model training.
• Legal challenges could reshape how AI companies access and utilize copyrighted material, impacting future innovation.

Top artificial intelligence companies, including Anthropic and OpenAI, are encountering increasing legal challenges as they push the boundaries of data usage for training their models. This comes as the industry reaches a “data frontier,” where accessible, unlicensed data is becoming scarce, leading to more aggressive data collection tactics and subsequent legal disputes.

Anthropic Sued for Alleged Copyright Infringement

This month, Anthropic, a San Francisco-based AI start-up, was sued by a group of authors who accuse the company of unlawfully using their copyrighted works. The lawsuit claims that Anthropic “never sought — let alone paid for — a license to copy and exploit the protected expression contained in the copyrighted works fed into its models.” This case adds to a growing list of copyright litigations in the AI sector, reflecting a broader tension between content creators and AI developers.

The legal action against Anthropic follows a high-profile case brought by The New York Times against OpenAI and Microsoft in late 2023. The Times alleges that these companies are profiting from “massive copyright infringement” by exploiting the newspaper’s content without proper authorization. Should The Times succeed in its lawsuit, it could open the door to similar legal actions against other companies in the AI space.

The Data Frontier and Its Challenges

The rapid advancements in AI over the past 18 months have led to a new challenge known as the “data frontier.” As AI companies strive to improve their models, they are running out of easily accessible, large-scale datasets. This scarcity is pushing companies to explore deeper into the web, purchase private data, or develop synthetic datasets. However, these methods bring their own set of challenges, including higher costs and ethical concerns.

Alex Ratner, co-founder of Snorkel AI, which specializes in building and labeling data sets, stated, “There’s no more free lunch. You can’t scrape a web-scale data set anymore. You have to go and purchase it or produce it. That’s the frontier we’re at now.” This quote underscores the growing difficulties AI companies face as they navigate this new phase in data acquisition.

Anthropic, which brands itself as a “responsible” AI company, has also been accused of “egregious scraping” of web data to train its systems. Similar accusations have been directed at Perplexity, an AI-powered search engine that competes with Google. These cases highlight the increasing scrutiny AI companies are under as they collect data for model training.

Struggles Between AI Start-Ups and Publishers

As AI start-ups race to develop more advanced models, they are encountering resistance from publishers and content creators who argue that their work is being unfairly exploited. The New York Times, in its case against OpenAI, argues that the AI company has cannibalized its content, using it in ways that directly compete with the newspaper and divert audiences away from it.

The outcome of The Times’s lawsuit could set a new legal precedent, determining how AI companies can use copyrighted material in their model training processes. Currently, AI companies like OpenAI and Anthropic are striking deals with publishers to access their content legally. OpenAI, for instance, has formed partnerships with Condé Nast, The Atlantic, Time, and The Financial Times to ensure that its models produce accurate and up-to-date responses.

Anthropic, however, has yet to announce similar partnerships, which may leave it vulnerable to further legal challenges. The company did recently hire Tom Turvey, a former Google executive with experience in managing partnerships with publishers, possibly signaling an intention to address these issues in the near future.

Legal Landscape for AI Companies

The legal environment surrounding AI companies and copyright is evolving rapidly. Google set an early precedent in 2015 when it won a case against a group of authors who claimed the company’s scanning and indexing of their works violated fair use. The court ruled in favor of Google, finding that the company’s use of the content was “highly transformative.”

However, The Times’s lawsuit against OpenAI challenges this notion, arguing that there is nothing “transformative” about how OpenAI uses the newspaper’s content. The outcome of this case could have far-reaching implications for how AI companies use copyrighted material in the future. 

Google’s victory took a decade to achieve, during which time the company solidified its dominance in the search engine market. AI companies today face a much faster-paced environment, where legal outcomes could significantly influence their ability to innovate and compete.