AIs can generate near-verbatim copies of novels from training data

Here's something that might surprise you — AI models can now produce near-perfect copies of bestselling novels, just from their training data. So, why does this matter? Well, it challenges the industry's claim that these systems don’t store copyrighted works. According to Melissa Heikkilä writing in the Financial Times, recent studies show that giants like OpenAI, Google, and Meta have a lot more of these works memorized than we thought. That’s a big deal because it means the classic defense — 'we don’t store copyrighted content' — starts to fall apart. Legal experts are warning that this could seriously impact ongoing copyright lawsuits, as it blurs the line between learning from data and copying it outright. Now, here’s where it gets interesting — if these models can generate verbatim copies, what does that mean for creators and publishers? As Heikkilä reports, we might need a whole new way to think about AI and intellectual property. Keep an eye on this space; it’s not going away anytime soon.