privacy

Pinterest Is Training AI on Every Photo You've Ever Pinned

Pinterest is training AI on every photo you've pinned — retroactively. An August 2026 class-action investigation is underway. Here's how to opt out.

You didn’t upload your photos to a training dataset. You saved inspiration for a home renovation, a wedding colour palette, or a dinner recipe. But as of April 2025, Pinterest updated its terms of service with language that makes every pin you’ve ever saved — going back to the day you joined — available to train its generative AI model, Pinterest Canvas.

You were opted in automatically. The opt-out is buried. And in August 2026, a law firm opened a formal class-action investigation into whether Pinterest obtained users’ informed consent before making this retroactive data grab.

What Pinterest Changed and When

Pinterest revised its privacy policy on April 30, 2025. The key clause reads:

“Pinterest may use information you’ve provided or that we’ve collected to train, develop and improve our technology such as our machine learning models, regardless of when Pins were posted.”

That bolded phrase — regardless of when pins were posted — is the part that matters. It means photos you saved or uploaded years before this policy existed are now covered by it. Pinterest didn’t ask for new consent for old content. It simply extended its rights backward through time.

The change covers Pins you’ve saved, photos you’ve directly uploaded, and Pin metadata including descriptions and tags. Pinterest trains its Canvas image-generation model on this material.

How Big Is the Training Set?

Pinterest Canvas, the company’s proprietary image-generation foundation model, was trained on approximately 500 million rows of publicly available Pins, their descriptions, and their metadata. That figure comes from Pinterest’s own disclosures.

Pinterest has over 550 million monthly active users and hosts more than 400 billion Pins. For scale: the training corpus that Pinterest built from user-generated content dwarfs what most companies are willing to disclose.

The Pins used for training come from content users have saved or uploaded — content for which Pinterest holds a broad license granted at sign-up. The updated terms made the AI training use of that license explicit, and retroactive.

The Class-Action Investigation

In August 2026, the law firm Migliaccio & Rathod LLP opened a formal investigation into Pinterest’s AI training practices. The firm is evaluating whether Pinterest’s use of user-uploaded content to train Canvas constitutes a violation of users’ reasonable expectations — and whether the opt-in-by-default structure meets the consent thresholds required under applicable law.

Class-action investigations at this stage do not represent a filed lawsuit. But law firms open them when they see a viable claim: a large user base, a clear policy change, and affected users who didn’t get meaningful notice.

This investigation mirrors the pattern seen elsewhere. Google faced scrutiny for using Google Photos content in AI training. Meta faced backlash for its changes to Instagram’s content terms. Amazon Photos faces ongoing litigation over facial recognition training on uploaded photos. Pinterest is now in that same company.

Why Your Saved Photos Are More Sensitive Than They Look

Pinterest boards feel casual. A recipe collection. A travel bucket list. A nursery design for a child you haven’t announced yet.

But the images people save to Pinterest reveal a detailed picture of identity, aspiration, health, and life stage. Someone creating a “postpartum recovery” board, a “AA anniversary gifts” board, or a “disability-accessible home design” board is creating a record of experiences they likely didn’t intend to contribute to commercial AI training data.

When those images are used to train a generative model, the patterns encoded in your curation — the visual aesthetic, the life context, the timing — become embedded in a model that Pinterest and potentially its commercial partners will use indefinitely.

How the Retroactivity Makes This Worse

Most data consent debates focus on what a company will do with new data going forward. The retroactive scope of Pinterest’s policy change is less common and more troubling.

The standard argument for retroactivity is that users granted a license when they signed up, and the company is simply exercising rights already held. But when a user saved a photo to Pinterest in 2014 or 2018, the use of that content for AI training was not a reasonable expectation — the technology didn’t exist at commercial scale, and Pinterest’s terms didn’t mention it.

Retroactive data rights expansions raise a genuine legal question: at what point does a service provider’s unilateral extension of how it uses historical data cross from “permissible under terms” to “unfair under consumer protection law”? The Migliaccio & Rathod investigation is trying to answer that question.

What Pinterest Actually Offers as an Alternative

Pinterest does provide an opt-out. It’s accessible through:

Settings → Privacy and data → Generative AI privacy settings → Toggle off “Use my data to improve Generative AI”

The toggle was not surfaced in a push notification, a pop-up, or an email when the policy changed. It was added to the privacy settings section, discoverable only if you went looking for it. Most users haven’t gone looking.

If you’ve never checked this setting, your pins from 2009 onward are in Pinterest’s training data right now.

What “Opt-Out” Means and Doesn’t Mean

Pinterest’s opt-out stops future use of your content for AI training — at least under the stated policy. It does not delete content already used in training runs before you opted out.

This is a structural limitation of all retrospective opt-outs from AI training. Neural network training is not reversible. Once a model has processed your content and updated its weights accordingly, those updates persist. Opting out prevents future training runs from including your data; it does not remove the influence your data has already had on the model.

The practical implication: opting out today is better than not opting out. But it is not a complete undo of the data use that has already occurred.

The Boarder Privacy Problem With Pinterest

Pinterest is notable as a platform where users curate visual content about other people — celebrities, public figures, unidentifiable models — but also, frequently, about themselves. Users pin photos of their own spaces, their children, their families.

When those photos are used to train facial recognition or image generation systems, the facial data embedded in them becomes part of the model. Pinterest’s Canvas terms are focused on image generation rather than facial recognition, but the visual geometry of people appearing in pinned images is still processed during training.

This is the same concern at the center of the Amazon Photos facial recognition litigation. The photos aren’t just artistic preferences — they contain biometric information about real people who never consented to contribute to a commercial model.

What Pinterest Doesn’t Clarify

Pinterest’s disclosures describe training on “publicly available” pins and pins “users have saved.” This language creates ambiguity: does training apply to private boards (boards you haven’t made public)?

Pinterest allows users to create secret boards, visible only to themselves. The privacy policy language doesn’t explicitly carve out secret board content from the AI training clause. If you’ve used private boards to save health-related, financial, or otherwise sensitive inspiration, the status of that content is unclear.

Pinterest has not issued a clear public statement distinguishing private board content from the AI training pipeline.

Who This Affects Most

Pinterest’s user base skews heavily toward visual curation for life planning: weddings, home design, parenting, health, fashion. These categories are precisely where saved images are most personally revealing.

If you’ve used Pinterest to plan anything meaningful — a medical procedure, a major life event, a private aspiration — the images associated with that planning are candidates for AI training, unless you’ve already opted out.

The scope of the opt-in-by-default decision, combined with the retroactive application, makes this a broader exposure than most users have registered.

What You Can Do Now

Opt out immediately. Settings → Privacy and data → Generative AI privacy settings → disable “Use my data to improve Generative AI.” This stops future training use.

Audit your boards. Review what you’ve saved, particularly on private boards. If boards contain health, financial, or family content you wouldn’t want in a training set, consider whether Pinterest is the right place for it.

Download your data. Pinterest’s data export tool lets you download your full archive before making changes. Access it via Settings → Security and Logins → Download your data.

Delete boards you no longer use. Content you’ve deleted is no longer on Pinterest’s servers. Opt-out covers future training; deletion removes the source material.

Reconsider what you save going forward. If you choose to stay on Pinterest, treating it as a public-facing curation tool — rather than a private archive — is the accurate mental model.

The Bigger Pattern

Pinterest is not unusual here. Over the past two years, Canva, Adobe, Google, Meta, Amazon, LinkedIn, and dozens of smaller platforms have all updated their terms to enable AI training on user-generated content. In most cases, users were opted in automatically, with opt-outs available if they went looking.

The cumulative effect is that a large portion of your digital life — photos you’ve saved, documents you’ve edited, messages you’ve sent — has been fed into AI models you didn’t consent to train.

Pinterest is the latest, and the retroactive framing makes it particularly clear about what this category of data use actually involves.


If you want a place to save personal content that doesn’t treat your files as AI training data, daftei starts with 5 GB free.

Your memories deserve better than an ad platform.

Try daftei free →
← All posts