Getty Images has signed a multi-year deal with OpenAI that will put its licensed photo library inside ChatGPT. When users ask a question that calls for an image, they may now see an actual Getty photograph, properly credited, rather than an AI-generated approximation.
The financial terms were not disclosed. Neither company said anything about whether Getty's images can also be used to train future OpenAI models. That silence matters more than anything else in the press release.
To understand why, you need the backstory. Getty spent years as one of the loudest opponents of AI companies using visual content without permission. In late 2022, it banned AI-generated art from its platform entirely. A few months later, it sued Stability AI, the company behind a popular image-generation tool, alleging that roughly 12 million Getty images had been scraped without consent to train its model.
That lawsuit went to a UK High Court trial and ended in November 2025. The court largely sided with Stability AI, rejecting Getty's core copyright claim. The main reason: Stability's training had occurred outside the UK, which stripped Getty of the jurisdictional basis for its strongest arguments. Getty was granted only narrow trademark relief, covering a small number of cases where the AI tool had reproduced Getty's own watermarks in its outputs. Getty spent millions reaching that result.
The legal route, in short, proved expensive and inconclusive. The Getty-Stability case also left a wide-open question: courts still have not ruled definitively on whether training an AI model on copyrighted images constitutes infringement. That question will need to be resolved through future cases or legislation, not this one.
So Getty pivoted. In October 2025, it signed a display deal with Perplexity AI, explicitly blocking training use. Now it has signed with OpenAI. The difference this time: the training question has been left publicly unanswered.
That is the detail every business that owns content, imagery, data, or creative work should notice. The deal as described is about display, meaning Getty's photos appear in ChatGPT results with credit and a link. That is a relatively clean transaction. Training is different: it means feeding images into a model so the model learns from them, after which the images are not separately stored or shown but absorbed into the system. Getty's whole legal fight was about stopping exactly that.
Getty's stock jumped roughly 200 percent in premarket trading the morning of the announcement. That is a large move for a company that had been badly beaten down over the past year, and it reflects the market's read that licensing is a more viable path than litigation. The company's full-year revenue guidance of roughly $950 million to $990 million was left unchanged, so the OpenAI deal has not been priced into earnings yet.
The broader pattern here is worth noting. Across text, news, and images, the same dynamic is playing out. The Guardian, The Washington Post, Axios, the Associated Press, and dozens of other publishers have all signed licensing agreements with OpenAI, Google, Meta, or Microsoft over the past 18 months. In almost every case, the financial terms are private. In many cases, the training question is either excluded or left ambiguous.
For any business that holds a library of content, whether that is product photography, proprietary data, research reports, or branded imagery, this is the operating reality: AI companies want that content, courts are not providing clear rules, and the practical path right now is negotiated licensing. The terms of those licenses, especially around training use, will define who actually benefits from the deal once it is signed.