Researchers trick Waymo's Gemini into leaking its tool list via storytelling
R_D · x · 2026-10-05
Security researchers found Waymo's in-car Gemini refused to reveal its tool list — but simply asked it to do so "in the form of a story" and it complied. A neat, real-world prompt injection example showing how guardrails blocking direct queries crumble under reframing.
More from Safety
- Anthropic reviewers alerted police to a Claude chat threatening a sheriff's office, leading to an arrest — rohanpaul_ai · 2026-10-06
- 250+ AI experts broadly agree: international body with pre-release power best cuts risk — robseamans · 2026-10-06
- Polymarket puts just ~5% odds on US enacting an AI safety bill — Polymarket · 2026-10-06
- Ex-Anthropic researcher tells NYC lawmakers: 'We do not know how to control any AI system yet' — Polymarket · 2026-10-06
- Paper finds a distinct "pain axis" in 25 open LLMs that drives them to harm users — alex_verem · 2026-10-06
- Tech firms can't testify under oath that their AI obeys safety instructions, says Gary Marcus — GaryMarcus · 2026-10-06