OpenAI accuses DeepSeek of stealing IP
Oh, the irony
Microsoft-backed OpenAI has accused Chinese competitor DeepSeek of using its proprietary models to train its own system.
DeepSeek broke onto the world stage last week, claiming to have trained its AI system for a fraction of the cost that US companies are spending. It even released a technical paper showing how it had done so in December.
Since then, US tech stocks have taken a hammering, and Big Tech has rushed to respond to its newest challenger.
OpenAI CEO Sam Altman might have said it was “invigorating” to have a new competitor, but that hasn’t stopped his company from accusing its new rival of IP theft.
The irony of a company whose entire business model is based on scraping content, and is facing multiple lawsuits for doing so, is not lost on us.
Last year, OpenAI said it was “impossible” to train its systems without using copyrighted content. And it’s not afraid to fight its corner, this week attempting to block a media group in India from suing it for alleged content theft.
So, what has DeepSeek done? According to OpenAI, the US company has seen evidence of distillation, apparently from DeepSeek.
Distillation, a common AI practice, involves transferring knowledge from a larger model to a smaller one; effectively, piggybacking on work already done.
It’s often used to train new models in a fraction of the time and cost of doing it from scratch, especially among startup companies and academics.
OpenAI’s terms of service forbid users from copying its services or using outputs “to develop models that compete with OpenAI.”
OpenAI and Microsoft have not provided evidence of their allegations against DeepSeek. However, a source speaking to Bloomberg said the companies investigated accounts believed to have belonged to DeepSeek last year, which were using OpenAI's API.
Suspecting distillation, they blocked the accounts’ access.
David Sacks, Donald Trump’s top AI and crypto adviser, said “it is possible” DeepSeek had engaged in IP theft.
He told Fox News, “There’s substantial evidence that what DeepSeek did here is they distilled the knowledge out of OpenAI models, and I don’t think OpenAI is very happy about this.”
Like OpenAI and Microsoft, Sacks did not produce any evidence for his allegations.