From Your Feed to AI Fuel
When you post a photo to a public social media profile, a blog, or a photo-sharing site, you're making it visible to other people. But you're also making it available to automated programs called web scrapers. AI development companies use these scrapers to hoover
up immense quantities of data from across the internet to build datasets. These datasets, some containing billions of images, are the lifeblood of generative AI models. They teach systems like DALL-E, Midjourney, and Stable Diffusion to recognize patterns, objects, and styles. The process is straightforward: an AI is shown millions of photos of cats, for instance, until it learns to generate a new, unique image of a cat on command. The same goes for human faces, artistic styles, and specific locations—all learned from publicly accessible images.
Is Public Really Fair Game?
This is where the debate gets heated. AI companies often argue that using publicly available data is legally defensible under doctrines like 'fair use' in the United States. The argument is that the use is 'transformative'—the photos aren't being resold, but are being used to teach a machine, which is a fundamentally different purpose. However, creators, photographers, and privacy advocates strongly disagree. They contend that posting a photo for human viewing does not imply consent for it to be used as free raw material for a commercial AI product. A number of high-profile lawsuits have been filed by artists and stock photo agencies like Getty Images, who argue that this mass scraping constitutes copyright infringement on a grand scale. These cases are testing the boundaries of existing copyright law, which was not written with generative AI in mind.
Consent by Default, Not by Design
A key part of the problem lies in the default settings of most online platforms. Users are automatically opted in to having their public content used for AI training. Companies like Meta, the parent of Instagram and Facebook, have updated their privacy policies to explicitly state they use public user content to train their AI models. While they may not use private messages, public posts, photos, and captions are considered fair game. To prevent this, a user must actively seek out and navigate often-buried privacy settings to opt out—if an opt-out option is even available. In regions with strong data protection laws like the European Union, companies are required to provide a way for users to object. However, in the U.S. and other regions, such options are often not offered, leaving users with little choice but to accept the terms or delete their accounts.
The Push for Transparent Notice
In response to growing public and regulatory pressure, the call for transparency is getting louder. Meaningful transparency isn't just a line buried in a 50-page terms of service agreement. Instead, it would involve clear, upfront notices that inform users exactly how their data will be used to train AI models. New laws are beginning to mandate this. For example, the EU AI Act includes transparency obligations requiring providers to disclose when content is AI-generated. In the U.S., California's Generative AI Training Data Transparency Act, effective in 2026, requires developers to publicly disclose details about the data used to train their models. This includes whether copyrighted or personal information was used. The goal is to shift the industry from a model of quiet, assumed consent to one of explicit, informed choice, where users have genuine control over their digital creations and identity.
