AI-resistant tools for creators: what they are and how they work
As generative AI systems become increasingly sophisticated, the creative community is responding with a new arsenal of software tools designed to resist the unauthorised ingestion of their work into AI training datasets. These tools represent a growing movement to reclaim control over digital content and challenge the unchecked data harvesting practices of AI developers.
The problem: unconsented AI training
Generative AI models, like those powering image generators and large language models, are trained on vast datasets scraped from the internet. This includes millions of artworks, photographs, writings and other creative outputs, often without the creators’ knowledge or consent. While some argue this falls under “fair use” many artists and rights holders see it as digital appropriation, especially when AI outputs mimic their style or reproduce fragments of their work
The technical response: Glaze and Nightshade
Two of the most prominent tools in this resistance movement are Glaze and Nightshade, developed by researchers at the University of Chicago:
- Glaze is a defensive tool that subtly alters the pixels of an image to confuse AI models attempting to learn an artist’s style. These changes are invisible to the human eye but disrupt the model’s ability to replicate the original aesthetic.
- Nightshade, on the other hand, is an offensive tool. It poisons training data by embedding misleading signals into images. For example, an image of a cow might be interpreted by an AI model as a handbag, leading to distorted outputs. This increases the cost and risk of training on unlicensed data, incentivising developers to seek proper licensing.
Both tools are free to use and designed to run locally, ensuring privacy and control for creators. They can be used independently or together, with Nightshade soon to be integrated into WebGlaze for streamlined protection.
The legal and governance response: spawning AI and opt-out registries
Beyond pixel-level defences, platforms like Spawning AI offer creators a way to opt out of future AI training. Their ‘Do Not Train’ registry allows individuals to flag their intellectual property, signalling to AI developers that their content is off-limits. While not legally binding, major players like Hugging Face and Stability AI have agreed to honour these opt-outs.
Spawning also provides tools like the ai.txt file (a machine-readable directive similar to robots.txt) that websites can use to communicate AI training preferences. Their browser extension helps users identify if content on a webpage has been included in public training datasets.
The metadata approach: Adobe’s content credentials
Adobe’s Content Authenticity Initiative introduces a metadata-based solution. Through its Content Credentials, creators can attach verified information, such as their name, social media and opt-out preferences, to their work. These credentials are durable, tamper-evident and backed up in Adobe’s cloud, making it easier to assert authorship and discourage unauthorised use.
Adobe’s tools also allow creators to proactively request that generative AI models do not use their content for training or inspiration. While this relies on voluntary compliance, it sets a precedent for transparency and accountability in digital content sharing.
The legal debate: fair use vs. market harm
The legal landscape remains murky. A recent critique of the U.S. Copyright Office’s stance on generative AI argues that training models on copyrighted works can be transformative and fall under fair use, especially when outputs are novel and not derivative. However, many creators and advocacy groups maintain that style mimicry and unlicensed use undermine their livelihoods and violate the spirit of copyright law.
Looking ahead
The emergence of these tools marks a pivotal moment in the digital rights movement. Creators are no longer passive participants in the AI revolution, they are actively shaping its boundaries. Whether through technical sabotage, metadata tagging or legal advocacy, the message is clear: consent matters.
As AI continues to evolve, so too will the tools and frameworks designed to protect human creativity. The challenge now is to ensure these solutions are widely adopted, legally recognised, and ethically respected. We watch with interest!
_______________________________________________________________________________________________
Disclaimer
This information is intended for general informational purposes only and does not constitute legal advice. We recommend seeking professional advice before taking any action on the information provided. If you would like to discuss your specific circumstances, please feel free to contact us on 0800 2800 421.
