Syntax hacking: Researchers discover sentence structure can bypass AI safety rules
New research offers clues about why some prompt injection attacks may succeed.
https://arstechnica.com/ai/2025/12/syntax-hacking-researchers-discover-sentence-structure-can-bypass-ai-safety-rules/?utm_brand=arstechnica&utm_social-type=owned&utm_source=mastodon&utm_medium=social
Discussion
Loading...
Semantic analysis has difficulty identifying steganography buried in poetry, artwork, incomplete or misspelled sentences & words, jokes, sarcasm, pidgin, dialects, slang, non-European languages like Hopi or Navajo, or non-biblical narrative stories or parables?