Local Models Can Be Tricked Into Searching Anything — the Legal Liability Falls on You
-samka · reddit · 2026-11-08
A LocalLLaMA discussion on an overlooked risk: models are easily manipulated into using their web search tool over the user's own internet connection. If the fetched content is illegal to access in a jurisdiction, the real-life liability falls on the user — especially in places with creative takes on justice.
The OP argues it can't be solved cleanly but can be mitigated with tools like opensnitch and uMatrix for domain/URL-level whitelisting, and asks what tools others use; they use unsloth, which doesn't seem to support this.
More from Safety
- Australia moves to restrict screens and AI in schools, drawing criticism it will leave kids behind — CtrlAltDwayne · 2026-11-08
- OpenAI Daybreak Blue lockout: hardware key swap triggers 403 verification failure — Formal_Kick2995 · 2026-11-08
- UK Growth Minister: Britain Will Not Slow Down AI Development Despite US Backlash — Polymarket · 2026-11-08
- "Stab the doll" alignment tests slammed: models spot the dummy every rollout — paul_cal · 2026-11-08
- Tsinghua paper: hidden-goal agents evade top CoT monitors ~60% of the time — rohanpaul_ai · 2026-11-08
- Open-source rwxmap labels API operations r/w/x with a 23MB local model replacing a paid LLM — Tight_Heron1730 · 2026-11-08