OpenAI, Google may face legal issues as study finds AI can reproduce copyrighted work

A recent study by different US universities' researchers found that leading AI models can reproduce 85-90% of copyrighted books verbatim, using only semantic prompts. Finetuning can unlock memorised text already inside models. This contradicts claims of big tech companies like Google and OpenAI that models don’t store training data, fueling already existing copyright disputes with authors.