The Problem: Unauthorised AI Training
Every time you post an image online—whether it’s a personal photo on Instagram or a piece of digital art on a portfolio site—it’s potentially vulnerable. Companies developing large AI models, like DALL-E, Midjourney, and Stable Diffusion, train their
systems on massive datasets of images scraped from the public web. This process, often called ‘data mining’ or ‘scraping’, typically happens without the consent of the original creator or subject. For artists, this means their unique style can be mimicked and replicated by AI. For everyday users, it raises significant privacy concerns about how personal photos are being used by corporations for profit. While the legal and ethical battles are ongoing, you don’t have to be a passive bystander. You can take active steps to make your images a much harder, less appealing target for these data-hungry algorithms.
Your First Line of Defence: Platform Privacy
The simplest and most effective barrier is a locked door. Public social media profiles are an open invitation for web scrapers. The first step is to review and tighten the privacy settings on all your accounts. - **Instagram/Facebook:** Set your profile to ‘Private’. This ensures that only approved followers can see your posts. While not foolproof against all forms of data collection, it stops the most common, large-scale public scraping bots. - **X (formerly Twitter):** Use the ‘Protect your Posts’ feature in your privacy settings. This functions similarly, making your tweets and the media within them visible only to your followers. - **Portfolio Sites:** If you’re an artist or photographer using a public portfolio, this option is less feasible. However, check if your platform offers any built-in protection against right-clicking or image saving. While easily bypassed by savvy users, these features can deter casual scraping. The real power for creators comes from more advanced tools designed specifically to disrupt AI models.
Advanced Protection: Image ‘Cloaking’ with Glaze
For artists and creators who need to keep their work public, a tool called Glaze is a game-changer. Developed by researchers at the University of Chicago, Glaze is a free application that adds a nearly invisible layer of changes to your artwork before you post it online. To the human eye, the image looks the same. But to an AI, these changes are deeply confusing. When an AI model tries to analyse a ‘glazed’ image to learn an artist’s style, it gets junk data. For example, it might learn to associate your distinct painting style with the look of abstract modern art. If enough artists use Glaze, it can effectively disrupt an AI’s ability to mimic specific human creators. You download the app, run your images through it on your own computer, and then upload the protected versions. It’s a powerful defensive move to protect your unique artistic signature.
Fighting Back: ‘Poisoning’ Data with Nightshade
If Glaze is a defensive shield, Nightshade is a poisoned dart. Created by the same team, Nightshade is an offensive tool that also alters your images in subtle ways. However, its goal is more aggressive: to corrupt the AI models that scrape it. When a model trains on a ‘poisoned’ image, it learns incorrect information. For example, an image of a dog might be tagged and altered in a way that teaches the AI that dogs look like cats. If a model ingests enough Nightshade-poisoned images, its results become chaotic and unreliable. A prompt for a ‘car’ might generate a cow; a request for a ‘fantasy castle’ might produce a picture of a toaster. This technique doesn’t just protect your single image; it actively sabotages the models that are training on scraped data without consent, making the entire practice more costly and less effective for the AI companies.
Putting It All Together: A Layered Strategy
No single method is a silver bullet. The best approach is a layered one that combines several tactics. First, decide what needs to be public and what doesn't. Use strict privacy settings for all personal content. For public-facing creative work, use tools like Glaze and Nightshade before uploading. Add visible watermarks to your images—while they can be edited out, they are another hurdle for scrapers. Finally, be mindful of the Terms of Service for any platform you use. Many sites include clauses that grant them broad licenses to use your content. By combining platform settings, specialised software, and conscious online habits, you can build a robust defence for your digital identity and creative work.
