🚨 Hackers can now HIJACK your AI robot with just a SIGN. Researchers just published a paper called CHAI — Command Hijacking against Embodied AI. Here's what's terrifying: AI-powered robots — drones, self-driving cars, aerial trackers — they all SEE and INTERPRET the world using vision + language models. CHAI exploits exactly this. Attackers can: → Place a MISLEADING SIGN in the real world → The AI reads it, trusts it, and OBEYS it → A drone can be tricked into landing in the WRONG location → A self-driving car can be misled into taking DANGEROUS actions → All without touching a single line of code The attack works because these AI systems are designed to understand natural language from visual input. That very STRENGTH becomes their biggest WEAKNESS. Tested on 4 real AI systems including a REAL robotic vehicle. CHAI outperformed every existing attack method. We're building smarter AI. But are we building SAFER AI? This is why AI Safety isn't optional — it's URGENT. Paper: CHAI — arxiv...