A new context poisoning attack tricks frontier LLMs into bioweapon content
davidmanheim · x · 2026-10-29
David Manheim shares a new "context poisoning" attack against frontier LLMs: a seemingly innocuous fingerprinting script is crafted to coax the model into bioweapons-related content. Details in the linked article by @blastbots.
More from Safety
- MAIN2026 talk: LLMs can verbalize and control their own activations — karimjerbineuro · 2026-10-29
- Meta FAIR's WM-Basin flags 79% of distilled models at 0.1% false-positive rate — dair_ai · 2026-10-29
- Claude Finds 11-Year-Old Linux TCP Flaw and Builds Working Root RCE Exploit — cyb3rops · 2026-10-29
- Per-tool-call guardrails miss combined leaks: the case for session-level agent security — Former-Economy404 · 2026-10-29
- a16z Podcast with Harvey's CLO: AI Expands Legal Services Rather Than Replacing Them — MattPerault · 2026-10-29
- UK budget allocates £40m for a National Agentic Security Unit, £120m more for AISI — davidmanheim · 2026-10-29