jleemiller.bsky.social
@jleemiller.bsky.social
Reposted
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."

https://arxiv.org/html/2511.15304v1
November 20, 2025 at 5:06 PM
2/1/2025
February 1, 2025 at 3:08 PM
Brunswick, Georgia
January 4, 2025 at 12:08 PM