Hey all. Sorry for a relatively long quiet period. A lot has been going on in 2025 and keeping us busy. Aside from the day job of research, there has been a lot of work with a few legal cases, and in recent months, much work writing and advocating for AI transparency laws. Time for an update.

Show Post

We have other research results which I'll announce once they're ready. Finally, I finally figured out some things on LLMs. We have 3 very new directions going beyond protecting visual art: a) a meaningful poison attack on LLMs, b) better ways to block AI scrapers, and c) some ideas on AI music.

Show Post

The work is peer reviewed and will be published at ACM CCS 2025 (another of the big 3 in computer security conferences). The paper is here: arxiv.org/abs/2506.21874. Like Nightshade, we are working on a binary implementation targeted for release late this summer.

Show Post

Update thread #2. Hemlock One thing we worked on is a different tool to complement Nightshade. We discovered a new attack on diffusion models that is more generalizable than Nightshade. It targets an entirely different part of AI training pipeline, basically making AI models poison themselves.

Show Post

If you're going to SDCC, come by and say hello. I'm also really looking forward to attending Lightbox again in October. We'll have another panel at LBX updating artists and all on technical, legislative, and legal efforts to protect creatives across different modalities.

Show Post