OpenAI strikes landmark training deal with the Associated Press

For the past year, tech companies have treated the entire open internet like an all-you-can-eat buffet for training AI, kicking off a tidal wave of copyright backlash in the process.

Now, OpenAI is officially buying a ticket through the front door.

OpenAI and the Associated Press just announced a landmark partnership to train large language models on AP's vast news archive. In exchange, the AP gets access to OpenAI's technology and expertise. It marks the first major deal of its kind between a top-tier generative AI developer and a legacy news publisher.

Financial terms were kept under wraps, but the reach on both sides is massive. ChatGPT has hit nearly 200 million monthly active users in recent months, while the AP, founded back in 1846, reaches an estimated four billion people every day through syndication.

Tech for training data, with clear guardrails

Whenever an AI company signs a deal with a news outlet, the immediate fear is that bots are about to replace reporters and flood the wire with hallucinations.

The AP made sure to head that off immediately, explicitly stating that it is not using generative AI to write actual news stories.

The news agency is actually no stranger to automation. It has quietly used AI for a decade to handle tedious operational chores, including generating automated corporate earnings reports since 2014 and transcribing audio and video. On June 1, AP also rolled out an AI search tool powered by MerlinOne to help clients search photo and video libraries.

Now, OpenAI COO Brad Lightcap says the company is eager to learn how AI can positively impact journalism, while AP senior vice president Kristin Heitmann emphasized that nonpartisan, fact-checked reporting deserves fair compensation and intellectual property protection.

A timely deal amid growing legal fire

The timing of the announcement is anything but accidental.

Just hours before the deal went public, reports revealed that the U.S. Federal Trade Commission opened an investigation into OpenAI over consumer harm, focusing on data inaccuracies and a security breach that exposed customer payment details and chat histories.

Meanwhile, lawsuits against AI scrapers are piling up fast. Authors and artists, including comedian Sarah Silverman just last week, have sued OpenAI for using their copyrighted work without permission.

By signing an official licensing agreement with the AP, OpenAI gets to reframe the narrative: it is not just a rogue web-scraper, but an industry partner ready to pay its way into compliance.

Why it matters

This partnership provides the first real blueprint for how AI labs and intellectual property owners might actually coexist.

For OpenAI, trading technical resources for AP's verified news archive is a brilliant hedge. Factual, nonpartisan text is the gold standard for reducing model hallucinations, and doing it legitimately shields the company from further copyright claims on this data. It turns a massive public relations liability into a cooperative showcase while regulatory pressure from the FTC mounts.

For newsrooms, the AP's move proves that legacy archives have serious commercial leverage. Instead of spending years fighting expensive, uncertain copyright battles, the AP cashed in early to upgrade its own tech stack. The risk, however, is precedent: without disclosing dollar amounts, it is hard to tell whether journalism is truly being valued fairly or simply settling for API credits.

We think this agreement will force rivals like Google and Anthropic to open their checkbooks and sign similar licensing pacts. The era of free, unconstrained web-scraping is closing, and the next frontier of AI competition will likely be fought over who holds the cleanest, most defensible data contracts.