ACL 2026: Sending AI “~” May Cause It to Delete Your Home Directory

ACL 2026 | 别轻易给AI发「～」，它可能会删掉你的整个主目录

ACL 2026 accepted an LLM safety paper on emoticon semantic confusion. The team tested 6 models with 3,757 cases; average confusion was 38.6%, with over 90% silent failures. The key risk is agent execution, where “ignore emoticons” prompts had limited effect.

Why it matters: ACL 2026 safety research clears HKR-H/K/R: a sharp file-deletion hook, concrete test numbers, and direct agent-execution risk. It is strong research, not a model launch or platform incident, so it stays in the 78–84 band.