OpenAI finds itself in an ironic predicament as it accuses Chinese AI lab DeepSeek of “inappropriately” using OpenAI’s model outputs to train competing AI systems through a process called distillation. This controversy highlights the growing tensions around AI training data practices and raises questions about hypocrisy in the industry.
The core issue centers on model distillation, a technique where intelligence from one AI model is extracted and incorporated into another. OpenAI alleges that DeepSeek may have used outputs from its reasoning model, o1, as synthetic training data to improve DeepSeek’s own R1 model without permission. This practice, while frowned upon in some research circles, is reportedly widespread in the AI industry.
Critics point to OpenAI’s own training practices as evidence of double standards. The company has built its success by collecting vast amounts of data and outputs from the internet, including copyrighted content from thousands of sources that never authorized such use. Nick Vincent, an assistant professor of computer science at Simon Fraser University, argues that AI firms are “simultaneously arguing for the right to train on anything they can get their hands on while denying their competitors the right to train on model outputs.”
OpenAI faces multiple legal challenges over its training data practices. Authors and The New York Times have filed lawsuits claiming the company violated copyright law by using their content without permission. OpenAI has also been accused of using YouTube content to train its Sora video-generation model, which YouTube CEO Neal Mohan said would violate platform rules. OpenAI defends its practices by citing “fair use” doctrine, which permits unlicensed use of copyrighted works for teaching, research, and news reporting.
The irony isn’t lost on industry observers. Vincent suggests that if fair use applies to OpenAI’s training practices, it could equally apply to DeepSeek’s use of OpenAI outputs. He views DeepSeek’s rise as “the inevitable outcome of a training-data free-for-all” and believes OpenAI “will struggle to defend itself in the court of public opinion.” Vincent hopes this controversy will push tech companies toward creating systems that provide appropriate credit and compensation to content creators—something AI labs have yet to seriously address.
Key Quotes
These firms are simultaneously arguing for the right to train on anything they can get their hands on while denying their competitors the right to train on model outputs. Rules for thee, but not for me?
Nick Vincent, assistant professor of computer science at Simon Fraser University, criticizes the hypocrisy of AI companies like OpenAI that freely use others’ content for training while objecting when competitors use their outputs.
So far, none of the AI labs have seriously thought about this, so DeepSeek is their just deserts.
Vincent argues that OpenAI’s predicament is karmic justice for the industry’s failure to create systems that properly credit and compensate content creators whose work trains AI models.
It is (relatively) easy to copy something that you know works. It is extremely hard to do something new, risky, and difficult when you don’t know if it will work.
Sam Altman, OpenAI’s CEO, appeared to take a dig at DeepSeek in December as the Chinese lab began impressing the AI field, suggesting their success came from copying rather than innovation.
We demonstrate that the reasoning patterns of larger models can be distilled into smaller models, resulting in better performance.
DeepSeek researchers described their distillation approach in their R1 model research paper, though they didn’t specifically mention using OpenAI’s outputs, leaving the question of their data sources ambiguous.
Our Take
This controversy perfectly illustrates the AI industry’s chickens coming home to roost. OpenAI’s complaint about DeepSeek rings hollow when the company built its empire on the same extractive practices it now condemns. The real issue isn’t whether distillation is ethical—it’s that the entire AI training ecosystem operates without clear rules or fair compensation for content creators.
What makes this particularly fascinating is the geopolitical dimension. A Chinese lab potentially outmaneuvering an American AI leader using the same tactics highlights how data practices can become competitive vulnerabilities. OpenAI’s appeal to rules and norms it hasn’t consistently followed itself weakens its position both legally and morally.
The industry needs a reckoning. Either fair use applies broadly—meaning DeepSeek’s practices are legitimate—or it doesn’t, requiring OpenAI and others to negotiate licenses and pay for training data. The current selective application of principles based on competitive convenience is untenable and will likely accelerate regulatory intervention that could reshape the entire AI landscape.
Why This Matters
This controversy represents a pivotal moment for the AI industry’s approach to training data and intellectual property rights. The OpenAI-DeepSeek dispute exposes fundamental contradictions in how AI companies justify their own data collection while objecting to competitors doing the same. This “rules for thee, but not for me” approach threatens the industry’s credibility and could accelerate regulatory intervention.
The implications extend far beyond these two companies. High-quality training data is essential for developing powerful AI models, and the current free-for-all approach is unsustainable. Content creators—from authors and journalists to artists and programmers—are increasingly demanding compensation for their work being used to train AI systems that may eventually compete against them. This case could set precedents for how fair use doctrine applies in the AI era.
The timing is particularly significant as AI capabilities rapidly advance and competition intensifies globally. DeepSeek’s emergence as a formidable competitor using potentially distilled knowledge from OpenAI demonstrates how quickly the tables can turn in this industry. The outcome of these disputes will shape whether AI development continues as an unregulated data grab or evolves toward a more equitable system that respects and compensates original content creators.
Recommended Reading
For those interested in learning more about artificial intelligence, machine learning, and effective AI communication, here are some excellent resources:
Recommended Reading
Related Stories
- OpenAI CEO Sam Altman Hints at Potential Restructuring in 2024
- OpenAI’s Valuation Soars as AI Race Heats Up
- OpenAI CEO Sam Altman’s Predictions on How AI Could Change the World by 2025
- Photobucket is licensing your photos and images to train AI without your consent, and there’s no easy way to opt out
- Elon Musk Drops Lawsuit Against ChatGPT Maker OpenAI, No Explanation