The rise of AI and generative AI in the creative industry has led to concerns over copyright and ethical use, as machine-learning models can be trained using scraped imagery from the web, often without artists’ or creators’ consent.
Ethical hack developer Aditya Bhatt has launched PixelPhantomX, an open-source tool that applies a number of defences to images to fool algorithms into classifying them as unsuitable for training, including adversarial noise injection, watermarks and edge distortion.
Artists can thus “poison the well” of AI datasets and protect their digital art from being copied or cloned by machines.
Bhatt’s paper, titled “Pixel Poisoning: Hacking Generative AI”, covers architectures including Glaze and Nightshade and encourages artists to fight back using open-source tools and techniques.