Monday, 28 April 2025
27.9 C
Singapore
27.4 C
Thailand
19.7 C
Indonesia
27.6 C
Philippines

OpenAI under fire as new study reveals signs of using copyrighted content in training

A new study suggests OpenAI’s models memorised copyrighted content, raising concerns about fair use and data transparency.

A recent study suggests that some of OpenAI’s AI models may have learned directly from copyrighted material — without permission. This finding adds weight to ongoing legal battles by authors, developers, and other rights-holders who say their work has been unfairly used to build these models.

While OpenAI has argued that using such content is covered by “fair use,” the study raises new concerns over whether that defence holds up. Researchers from the University of Washington, Stanford University, and the University of Copenhagen worked together to develop a method for checking whether AI models have memorised specific pieces of text during training.

New method reveals hidden memorisation

The study focused on what the authors call “high-surprisal” words — uncommon words that stand out when placed in certain sentences. For example, “radar” in the sentence “Jack and I sat perfectly still with the radar humming” is considered a high-surprisal word. It’s less expected in this context than words like “engine” or “radio,” which might appear more often before the word “humming.”

Using this idea, the researchers tested several of OpenAI’s models, including GPT-3.5 and GPT-4. They took text snippets from fiction books and articles published in The New York Times, removed the high-surprisal words, and asked the models to guess the missing word.

If a model guessed the correct word with high accuracy, it suggested that the model had seen that exact phrase or passage during its training — an indication of memorisation. Since the data included copyrighted books and journalism, this poses serious ethical and legal concerns.

GPT-4 showed signs of memorising books

The tests showed that GPT-4 — OpenAI’s most advanced model — appears to have memorised sections from popular fiction. Some of this content came from a dataset called BookMIA, which includes samples from copyrighted ebooks. The model also seemed to recall parts of New York Times articles less frequently than it did with fiction.

These findings point to the possibility that GPT-4 was trained, at least in part, on copyrighted materials. That’s a major issue for creators whose work may have been included without their consent.

Abhilasha Ravichander, a PhD student at the University of Washington and one of the study’s authors, explained the significance of this discovery. “To have large language models that are trustworthy, we need to be able to audit them and understand how they work,” she said. “Our study offers one way to investigate that, but it also highlights the urgent need for more transparency around the data these models are trained on.”

OpenAI has been the subject of several lawsuits over its use of copyrighted content. It has defended its approach by arguing that training AI with such content is fair use—a legal principle in the US that allows for limited use of copyrighted material without needing permission.

At the same time, OpenAI has tried to show it takes content rights seriously. It has licensing agreements with some publishers and offers an “opt-out” process so creators can request that their work not be used in training.

Still, the company continues to lobby governments around the world to support looser rules regarding AI training. It wants clearer legal protections that would allow models to be trained on a broad range of online content—including some copyrighted material—without facing legal risks.

But as this new study shows, there’s a fine line between learning from data and copying it. And until lawmakers draw that line, the debate around fair use in AI training will likely remain heated.

Hot this week

Global PC shipments rise 6.7% in early 2025 as AI and tariffs drive demand

PC shipments rose 6.7% in Q1 2025, boosted by AI demand and tariff concerns, but growth is expected to slow later in the year.

AMD optimizes models for faster generative AI performance with Amuse 3.0

AMD announces optimised models for faster AI performance with Amuse 3.0, featuring enhanced photo filters, video diffusion, and more.

Anthropic aims to uncover how AI models think by 2027

Anthropic CEO Dario Amodei aims to understand how AI models work by 2027 and urges industry-wide action for safety and transparency.

Netflix raises subscription prices in Singapore again

Netflix again raises subscription prices in Singapore, with new rates for all plans and extra member slots.

StarHub celebrates 25 years of connection and innovation

StarHub celebrates 25 years of connecting Singapore, marking the milestone with island-wide festivities, giveaways, and new entertainment experiences.

Gitex Asia x Ai Everything Singapore highlights robotics, AI and next-gen tech at inaugural event

Gitex Asia x Ai Everything Singapore highlights robotics, AI, startups, and tech innovations, shaping Southeast Asia’s digital future.

Lenovo introduces new ThinkPad mobile workstations and business laptops for the AI-ready workforce

Lenovo refreshes its ThinkPad lineup with new AI-ready mobile workstations and business laptops, enhancing mobility, performance, and security.

SquareX secures US$20 million to transform browser security

SquareX raises US$20 million to strengthen browser security, offering enterprises an easy way to protect users without disrupting their workflows.

Smart Communications acquires Joisto to strengthen cloud archival capabilities

Smart Communications acquires Joisto to expand cloud-based customer conversation and archival solutions, strengthening its leadership in CCM and IXM.

Related Articles

Popular Categories