Our approach to data and AI

OpenAI dropped a blog post this week that reads less like a product launch and more like a statement of principles. A little over a year after ChatGPT went live and rewired how millions of people work, write and search for answers, the company is trying to get ahead of a question it can no longer dodge: where does all the training data actually come from, and who gets a say in it.

The centerpiece is something called Media Manager, a tool OpenAI says it's building for creators and content owners. The pitch is straightforward on paper — give rightsholders a way to identify their work and control whether it gets used to train future models. That's a meaningful shift in tone from a company that spent 2023 mostly fielding lawsuits and angry open letters from authors, news organizations and artists rather than offering them a dashboard.

There's not much technical detail yet on how Media Manager will actually work, which is worth flagging. OpenAI says it's coming, but a tool that lets creators flag and exclude their content is only as good as its accuracy and its reach — does it cover text, images, audio, all of it? Does it apply retroactively to models already trained, or only going forward? The post doesn't say, and that gap is exactly where skepticism tends to live.

Still, the timing tells its own story. This lands amid mounting legal pressure — The New York Times' lawsuit against OpenAI and Microsoft is the highest-profile example, but hardly the only one. Regulators in the EU and elsewhere are also circling the same territory: consent, compensation, and whether scraping the open web counts as fair use or just taking. OpenAI framing this as 'our approach' rather than a one-off fix suggests the company expects data provenance to be a permanent fixture of the conversation, not a controversy it can wait out.