Requesting Removal of Your Photos and Name From AI Training Datasets and Model Cards

Your photos and name can end up in public training datasets and model documentation without you realizing it. That exposure can lead to facial recognition matches, doxxing, or persistent search results tied to your identity. This guide explains how to locate where your information appears, understand your rights, and submit effective, respectful removal requests to dataset maintainers, hosting platforms, and downstream projects.

What This Means and Why It Matters

Training datasets often include images, captions, names, usernames, URLs, and metadata scraped from public sources. Model cards and data cards sometimes list contributors, subjects, or sources by name. Even if your social profiles are private today, past public content or third-party posts may have been captured and redistributed. Removing your information reduces biometric, reputational, and identity risks, and can minimize how often your image and name are resurfaced in new projects.

Common Places Your Photos and Name May Appear

• Public image datasets: Face datasets, person re-identification sets, and general image collections with captions or alt text that include names, usernames, or URLs.
• Academic and research repositories: Project pages or downloadable archives linked from universities, labs, or preprint papers.
• Open-source hosting: Platforms that host datasets and code, including release assets, data cards, and issue trackers that mention individuals by name.
• Model cards and data documentation: Pages that describe dataset sources, collection methods, licenses, and sometimes named examples or attributions.
• Mirrors and forks: Copies of datasets and their documentation on multiple sites or user accounts, sometimes with stale versions.

Before You Request Removal: Gather Evidence

Effective takedowns start with clear evidence. Document exactly where and how your data appears, and identify who’s responsible.

1. Search for references: Use your full name, common variations, usernames, and image-reverse searches. Include keywords like “dataset,” “model card,” “data card,” “image list,” or “annotations.”
2. Collect URLs and screenshots: Capture direct links to files, pages, and repository commits, plus screenshots showing your photo or name in context.
3. Note identifiers: Record file names, sample IDs, hash values, or index numbers used in annotations. These help maintainers find the exact entries to delete.
4. Check licensing and consent: Look for data licenses, consent statements, or IRB/ethics notes. If consent wasn’t obtained or the license doesn’t allow redistribution, mention that in your request.
5. Identify controllers and hosts: List the dataset maintainer (the entity that curates/releases it) and the hosting platform (the site storing it). You may need to contact both.

Know Your Rights and Angles

Removal pathways depend on jurisdiction and the type of information:

• Biometric and likeness concerns: Face images may fall under biometric or sensitive personal data. Emphasize risks of facial recognition and profiling.
• Personal data regulations: If you are in (or data was collected from) regions with privacy laws such as GDPR in the EU/UK or state laws in the U.S., you may have rights to deletion, correction, or objection to processing.
• Copyright or publicity rights: For photos you created, you may assert copyright. For recognizable images of you, some regions recognize rights of publicity or likeness rights.
• Platform policies: Hosting platforms usually have terms against doxxing, sensitive data exposure, or unauthorized personal data sharing. Policy violations can trigger removals even when law is unclear.

How to Contact Dataset Maintainers

Start with the entity that created or released the dataset. They can update the dataset, publish a new version without your data, and notify mirrors or downstream users.

1. Find the right contact: Look for an email in the dataset readme or data card, a “Contact” or “Maintainer” field, or a linked lab website. If missing, check the repository owner’s profile.
2. Use a clear subject line: Example: “Removal Request: Photo(s) and Name of [Your Name] in [Dataset Name], Entry IDs [IDs].”
3. Provide precise details: Include URLs, IDs, screenshots, and the exact data to remove (images, captions, annotations, names in docs).
4. Explain the basis: Briefly state privacy, consent, or legal grounds (e.g., no consent for biometric data; personal data rights request; copyright of your original image).
5. Request specific actions: Ask for deletion/redaction, an updated dataset release, purging of caches, and notification to known mirrors or users.
6. Set a reasonable timeline: Propose 14–30 days for response and action, and invite a confirmation once completed.

Template: Initial Maintainer Email

Subject: Removal Request: Photo(s) and Name of [Your Name] in [Dataset Name], Entry IDs [IDs]

Hello [Maintainer/Team],

I found my personal information in [Dataset Name] hosted at [URL]. The following entries contain my image/name:

• Entry/File IDs: [list]
• Direct URLs: [list]
• Screenshots: [attached or link]

I did not consent to inclusion of my image/name. This material constitutes personal data and creates biometric/identity risks. I request that you:

1. Remove my image(s), name, and related annotations from the dataset and documentation (including model/data cards).
2. Publish an updated dataset release without these items and note the change log.
3. Request removal from known mirrors/forks and downstream distributions under your control.
4. Purge cached copies on your hosting platform if applicable.

If relevant, I also assert the following basis: [GDPR/CCPA/biometric-sensitive data/copyright or likeness rights], and request deletion under applicable law.

Please confirm receipt and provide a timeline for resolution. I am happy to verify identity if required to prevent fraudulent requests.

Thank you,
[Your Name]
[Contact Email]

Requesting Redaction From Model Cards and Documentation

Model cards and data documentation may name you directly (e.g., in examples, attributions, or acknowledgments) or indirectly (links to your profiles). Even if images are removed, your name might remain in text.

• Ask maintainers to redact your name and identifiers and replace examples with non-identifying placeholders.
• Request updates to all versions of the documentation, including PDFs, releases, and site mirrors.
• Suggest a changelog note (e.g., “Removed personally identifiable information at the request of the individual”).

Escalating to Hosting Platforms

If maintainers do not respond or refuse, escalate to the hosting site using their abuse, privacy, or DMCA channels.

1. Locate the policy page: Find “Report content,” “Privacy complaint,” or “Takedown” instructions. Many platforms accept privacy removals beyond copyright claims.
2. File a detailed report: Include your evidence bundle, note sensitive personal data exposure, and link to specific files/commits/releases.
3. Cite applicable laws/policies: Reference local privacy rights if relevant, and the platform’s own rules against exposing personally identifying or sensitive data.
4. Track the case: Save ticket numbers, auto-replies, and response deadlines. Follow up if timelines lapse.

Handling Mirrors, Forks, and Caches

Once the original is updated, copies can linger.

• Request maintainer outreach: Ask the original maintainer to notify known mirrors and downstream users in release notes or mailing lists.
• Contact major mirrors directly: Use your evidence bundle and reference the updated source version that no longer contains your data.
• Ask platforms to purge caches: Some hosts cache large files or renderings. Request a reindex or cache clear.
• Monitor search engines: After successful removals, submit URL removals for outdated search results via search-engine tools when available.

Verifying Removals

Confirm the following after you receive a resolution notice:

• Your images, name, and annotations are no longer present in the dataset index or download archives.
• Model/data cards and readme files no longer reference your name, usernames, or links.
• New version numbers/releases clearly exclude your entries, and links to prior versions are taken down or gated.
• Major mirrors have synchronized with the corrected version.

If Your Data Is Already in a Released Model

When a model has already been trained, removing source data won’t retroactively delete its influence. Still, you can mitigate exposure:

• Request removal of your data from future dataset releases and documentation to prevent further propagation.
• Ask for model card updates acknowledging removal of your personal data from sources and noting any steps taken to reduce downstream risk.
• Request removal of example prompts/outputs that mention you, and seek redaction in demos or hosted inference apps.
• Consider platform-specific complaints if hosted demos generate content using your name or image in harmful ways.

Identity and Safety Considerations

Some requests require identity verification to prevent fraud. Share the minimum necessary details, and avoid sending full IDs unless the platform’s secure process requires it. If harassment or doxxing is involved, document threats and consider reporting to local authorities or platform safety teams.

Keep Organized: Your Removal Tracker

• Evidence bundle: URLs, IDs, screenshots, hashes.
• Contacts list: Maintainers, lab admins, platform trust/safety emails, abuse forms.
• Timeline log: Dates sent, responses, promised actions, version numbers, mirror updates.
• Resolution proofs: Final links, release notes, and confirmation emails.

Request Templates You Can Adapt

Short Follow-Up

Hello [Name/Team], following up on my removal request sent on [date] regarding [dataset/model card]. Could you share an update or estimated resolution date? Thank you.

Platform Escalation Summary

Hello Trust & Safety, I’m reporting exposure of my personal data in [dataset/repo URL]. It contains my image/name at [specific links]. I did not consent to inclusion. This creates biometric and identity risks. The maintainer has [not responded/declined]. I request removal of the referenced files and redaction of my name from documentation under your policies on personal data/sensitive information. Evidence attached. Thank you.

Reduce Future Exposure

• Lock down public profiles: Remove or privatize old albums and tagged images. Adjust who can download or index your photos.
• Watermark and license intentionally: If you publish images, use visible watermarks and explicit licenses that prohibit dataset use.
• Opt out where offered: Some projects provide formal opt-out processes; look for “Data subject rights,” “opt-out,” or “privacy request” links.
• Monitor mentions of your name and images: Set search alerts for your name, usernames, and unique phrases that appear in your captions or bios.
• Watch for identity misuse: Unusual credit activity can signal broader exposure from data scraping and breaches. Consider enrolling in privacy-focused credit and identity monitoring to catch and respond to risks early. One option is SmartCredit for privacy, credit monitoring, and identity protection.

FAQ

Will removal break the dataset?

Removing a few entries rarely harms overall utility. Responsible maintainers can update indices and release notes to keep versions consistent.

What if the dataset claims it only uses “public” data?

Public does not mean consequence-free. Many privacy laws still protect personal data gathered from public sources, and platform policies may prohibit redistribution of sensitive information without consent.

Can I force updates to downstream users?

You typically can’t compel every downstream user, but if the original dataset and major mirrors remove your data, many downstream copies will gradually age out. Platform takedowns can accelerate this.

Is a legal demand necessary?

Often no. A clear, respectful request with evidence works surprisingly well. If ignored, escalate through platform channels or seek legal advice for your jurisdiction.

Conclusion

Having your photos and name in training datasets or model cards can quietly expand your digital footprint. You can take control by identifying where your information appears, sending precise and well-supported requests to maintainers, escalating to hosting platforms when needed, and tracking mirrors and updates until the changes stick. With steady follow-up and simple monitoring habits, you can meaningfully reduce exposure and the risks that come with it.