OpenAI is building a Media Manager for creators

For over a year, AI labs and content creators have been locked in a bitter copyright stalemate. Creators argue their life's work was vacuumed up without permission, while AI companies point to fair use and vague web crawler standards.

Now, OpenAI is trying to build an actual off-ramp. On May 7, the company announced it is developing Media Manager, a new tool designed to let creators and content owners identify their works and choose whether they are included in or excluded from future model training.

OpenAI plans to launch the tool by 2025, promising it will handle text, images, audio, and video across multiple sources.

Beyond the limits of robots.txt

Until now, the only real lever for publishers has been robots.txt. OpenAI started honoring web crawler opt-outs last summer, but basically admitted that solution is broken for individual creators.

Most independent artists and writers don't own the servers hosting their work. Content gets remixed, reposted, and quoted across thousands of different websites, meaning a standard web crawl will swallow it regardless of the creator's wishes.

OpenAI says Media Manager will use cutting-edge machine learning to identify copyrighted media across various domains and enforce owner preferences. The company is consulting with creators, content owners, and regulators to shape the system.

Alongside the tool announcement, OpenAI detailed where its training data actually comes from. It leans on public web data (excluding known paywalls and personal identifiable information), proprietary partnerships like its deals with the Financial Times and private video archives for Sora, and opt-in user feedback. Business data from enterprise and API customers remains strictly off-limits.

OpenAI also pushed back on accusations of digital plagiarism. The company maintains its models learn statistical relationships rather than storing database copies, claiming instances where a model spits out verbatim content are considered a machine learning failure usually caused by heavily repeated web text.

Why it matters

Media Manager looks like a savvy political maneuver by OpenAI, but the devil will be in the execution.

For creators, having a granular opt-out mechanism across images, video, and audio sounds like a win. But it also shifts the burden onto the victim. Expecting millions of independent artists to manually register their catalogs across a proprietary OpenAI portal is asking a lot, especially if Google, Meta, and Anthropic don't adopt the exact same portal.

For the broader industry, OpenAI is trying to set the de facto copyright standard before regulators force one on them. If Media Manager becomes the benchmark, it creates an enormous compliance barrier that open-source developers and cash-strapped AI startups simply cannot afford to build.

There is also the elephant in the room: 2025. By the time this tool launches, frontier models will have already ingested the public internet several times over. An opt-out tool for future training does nothing to un-train the data already baked into GPT-4. It is a proactive peace offering, but we doubt it will stop ongoing copyright lawsuits in their tracks.