I've built a new method for steering LLMs called Semantic Overlays, small trained adapters on a frozen model which change how it perceives a piece of its context. The most readily applicable usage is to mitigate prompt injection, and it lets us take a very-injectable Qwen-3.5-9B to SOTA scores on all the prompt injection benchmarks I could find. (They are only blackbox attacks, but I did NOT train on anything like them — whitebox attacks are out of scope for this paper)
I'm excited for you to play with the tech — see if YOU can break it! (let me know if you can)
Paper at https://arxiv.org/abs/2608.23873 if you want to read more about it, code at https://github.com/JoshuaSP/semantic-overlays, adapters at https://huggingface.co/joshuapenman/semantic-overlays-adapte...
Also https://x.com/joshua_s_penman/status/2094823990472884389 if you wanna watch a little video I made!
Show HN: Semantic Overlays – an NX bit for LLM prompt injection (live demo)
semantic-overlays.vercel.app