DeepSeek accused of training on Google Gemini data
Researchers have reportedly found evidence that DeepSeek's R1 model was trained using data from Google's Gemini 2.5 Pro, based on similarities in language patterns and reasoning traces. DeepSeek and Google have not publicly responded to the allegations.
What happened
Researchers Sam Paech and an anonymous creator of SpeechMap found that DeepSeek's R1-0528 model employed words, expressions, and reasoning patterns similar to those of Google's Gemini 2.5 Pro, raising suspicions that DeepSeek used Gemini data for training.
DeepSeek did not reveal the specific sources of data used to train its R1 model. Google and DeepSeek have not publicly responded to the allegations.
Reported impact
The suspected misuse raises ethical and legal concerns regarding data use in artificial intelligence and intellectual property practices. This follows previous accusations that DeepSeek employed distillation techniques using OpenAI data, which may violate service terms when data is used outside authorized platforms.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.