Businesses and government agencies need a way to authenticate the images they exchange electronically. A classic solution is to hide a “digital watermark” in a color or gray-scale image by making imperceptible changes in the colors or brightnesses of individual pixels. But for black text on a white background, this approach doesn’t work so well: flipping even one white pixel to black on a field of white can produce a visible irregularity. Now University of Maryland information scientist Min Wu is perfecting software that adds secret data to such documents by subtly altering pixels in small blocks along the edges of letters, where the changes are virtually unnoticeable. Each block encodes a single binary digit; together, the digits constitute a signature that can be extracted to prove an image hasn’t been forged or altered. If it has, the hidden signature doesn’t appear, or becomes mush. Wu says officials at Princeton University, where she began work on the software as a PhD student, are talking with potential licensees about putting it to work in commercial document-verification systems.
This new data poisoning tool lets artists fight back against generative AI
The tool, called Nightshade, messes up training data in ways that could cause serious damage to image-generating AI models.
The Biggest Questions: What is death?
New neuroscience is challenging our understanding of the dying process—bringing opportunities for the living.
Rogue superintelligence and merging with machines: Inside the mind of OpenAI’s chief scientist
An exclusive conversation with Ilya Sutskever on his fears for the future of AI and why they’ve made him change the focus of his life’s work.
How to fix the internet
If we want online discourse to improve, we need to move beyond the big platforms.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.