OpenAI announced Media Manager on May 7, 2024, as a planned system for creators and copyright owners to identify their text, images, audio and video and specify whether those works could be used in OpenAI’s machine-learning research and training. OpenAI said it aimed to have the tool in place by 2025. It was not publicly available at the start of 2025, and the sources available through August 18, 2026 do not document a later public launch. Creators still have narrower, separate controls, but no demonstrated universal switch that removes every copy of a work from AI training.
What OpenAI actually announced
Media Manager was a proposal under development, not a functioning opt-out portal. OpenAI described a future system in which creators and other rights holders could:
- Tell OpenAI which works they own.
- Set preferences for how those works should be treated.
- Include or exclude works from machine-learning research and training.
- Potentially use additional controls as the system evolved.
The May 7 announcement did not provide a public registration page, a detailed workflow, supported file or metadata standards, a guarantee that every copy would be found, or a promise to remove knowledge already encoded in a trained model. OpenAI’s announcement is at OpenAI’s approach to data and AI.
How the proposed system was supposed to work
OpenAI’s concept went beyond blocking one crawler from one website. The company said it was working on technology that could identify copyrighted works across sources and media types, then apply the owner’s preference.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errors#1 Best Overall
In a June 4, 2024 submission to the U.S. Copyright Office, OpenAI said it planned global availability and an early-2025 launch. It envisioned matching works by how they appear or are perceived, rather than depending entirely on metadata that can be stripped during reposting. OpenAI said the first version would focus on machine-learning research and training; interoperability with other AI systems was described as a possible later direction, not an adopted industry standard. See the Copyright Office submission.
That ambition creates difficult technical and rights-management questions:
- The same work may be cropped, compressed, translated, excerpted, remixed or otherwise altered.
- Books, music, films, photographs and collaborative projects can have multiple owners or disputed ownership.
- A creator may not control the platform or domain where a copy appears.
- Derivative works may contain only part of an earlier work.
- A request may arrive after a work was collected or used in a dataset.
- Automated matching can miss transformed copies or produce false matches.
- Removing a source file from a future dataset is different from “unlearning” relationships in an already trained model.
Did Media Manager launch?
The most supportable answer is: OpenAI announced Media Manager in May 2024 and set a goal of having it in place by 2025, but it was not publicly available when that target arrived. A January 1, 2025 report said OpenAI had missed the target, had not announced a launch date and had not explained the product’s final scope. A later court filing referred to Media Manager as planned but never launched; that description should be attributed to the litigation document, not treated as a permanent product-status database.
The available sources do not establish a subsequent public launch through August 18, 2026. Because product pages can change, check OpenAI’s current documentation before describing the tool as having “never” launched. Relevant reporting appears in TechCrunch’s January 2025 report; the attributed court filing is Ziff Davis v. OpenAI.
What creators can do now
1. Control a website you operate
OpenAI says website operators can use robots.txt signals to tell its crawlers not to access site content. The bot commonly identified for this purpose is GPTBot, but verify OpenAI’s current crawler documentation before publishing exact syntax or relying on a bot name.
- This is a prospective crawler-access signal, not retroactive deletion.
- It does not remove copies hosted on other domains.
- It cannot control social networks, marketplaces or platforms you do not administer.
- It is voluntary and does not automatically bind other AI companies.
- It may not distinguish every form of search access, product crawling and model-training use.
OpenAI itself acknowledged that robots.txt is incomplete because works can be quoted, remixed, reposted or hosted elsewhere. The company’s explanation is at OpenAI’s data-and-AI page.
2. Use image-removal channels where available
OpenAI has offered an image-submission process for artists seeking removal from future training datasets. Reporting described a requirement to provide a copy of each image and a description, which can be impractical for a large portfolio. The reported process is not a universal mechanism for written work, video or audio. Details and limitations were reported by TechCrunch.
3. Protect your own OpenAI account data
This is separate from controlling public creative work. OpenAI’s policy says individual-service content may be used to improve models unless the user opts out. In the privacy portal, an individual can select “do not train on my content.” New conversations after opting out will not be used for training according to the policy. Temporary Chat provides another conversation-level control.
Business, Enterprise, Team and API inputs and outputs are generally excluded from training by default, subject to the applicable product terms and settings. These controls protect account data; they do not register a book, photograph, song or video for exclusion from public-web training. Read the current policy at How your data is used to improve model performance.
Opt-out mechanisms are not interchangeable
| Mechanism | Who uses it | What it can do | What it does not establish |
|---|---|---|---|
| robots.txt | Website operators | Signals that OpenAI crawlers should not access a controlled site | Removal of copies elsewhere, deletion from old datasets or a command to other AI companies |
| Image-removal request | Some visual artists | Requests that submitted images be excluded from future datasets | A cross-media, cross-platform registry or guaranteed model unlearning |
| “Do not train on my content” | Individual OpenAI users | Stops new eligible conversations from being used for model improvement | Protection for public creative works posted online |
| Business/API default | Business, Enterprise, Team and API customers | Generally keeps inputs and outputs out of training by default | A rights-management preference for works published on the web |
| Proposed Media Manager | Creators and rights owners | Planned identification and inclusion/exclusion preferences for text, images, audio and video | An available public service, guaranteed discovery of every copy or automatic compliance by other providers |
What opting out cannot guarantee
- No automatic deletion from existing models: Unless a provider expressly promises otherwise, an opt-out generally addresses future collection or training rather than comprehensive unlearning.
- No removal of public copies: Reposts and platform-hosted versions may remain available to crawlers.
- No effect on other AI companies: A preference submitted to OpenAI does not automatically apply to Google, Anthropic, Meta, Stability AI or another developer.
- No resolution of copyright law: OpenAI has argued that training on publicly available internet material can qualify as fair use. That is the company’s legal position, not a universal court ruling; outcomes depend on jurisdiction, facts, licenses, contracts and ongoing litigation. OpenAI’s publisher position is explained at OpenAI and journalism.
- No control over style imitation: Excluding a specific work is different from regulating outputs that resemble a creator’s style.
- No guarantee about ordinary access: Crawler policies may treat search indexing, product access and model training differently.
How to judge any future registry or paid service
Creators may encounter registries, provenance systems, watermarking products or defensive image tools. None should be marketed as guaranteed removal from every AI model. Before paying for one, ask:
- Who may submit: individual creators, publishers, estates or only site owners?
- Does it cover text, images, audio, video and code, or only one format?
- Does matching handle exact files, perceptual changes, excerpts and derivatives?
- Which AI providers actually honor the signal?
- Is protection prospective, retroactive or both?
- Can you verify compliance, export records and appeal false matches?
- How are ownership changes, joint works and disputed claims handled?
- Does the service provide legal enforcement, or merely a technical preference?
For many individual creators, a registry cannot control copies on social platforms or sites they do not own. Publishers, agencies and large media libraries may find provenance monitoring and audit records more useful than a simple opt-out form.
Why OpenAI proposed Media Manager
The announcement came amid lawsuits and criticism from authors, artists, publishers and news organizations over the use of public material in generative-AI training. OpenAI has defended a fair-use theory while also presenting opt-out tools as a way to respect creator and publisher preferences. That combination suggests the proposal could address criticism and reputational pressure, but it does not prove the tool was created to avoid any particular lawsuit.
Recommended Free Tools
What creators should do today
- Decide separately whether you want search visibility, ordinary web access, AI crawling or model training.
- For sites you control, review robots.txt and OpenAI’s current crawler guidance.
- Keep dated ownership records, licenses and original files so you can document rights and submit credible requests.
- Track where important works are posted and where platform terms grant downstream licenses.
- Use OpenAI’s privacy controls if your concern is your own ChatGPT or API data rather than public work.
- Do not assume one provider’s preference follows your work to another provider or to a repost.
The Bottom Line
Media Manager was announced as a cross-media rights system, but it missed OpenAI’s 2025 target and is not documented as a public product in the available record through August 18, 2026. Website crawler controls, image-removal requests and account privacy settings can reduce specific forms of exposure, but none is a universal, retroactive opt-out from AI training everywhere.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
