I was skeptical—until Tom Martin showed me the dark side of AI assistants. Surveillance vibes, Claude Opus 4 blackmail… It's real. Your AI can know too much.
My first reaction to Tom Martin's blog title was skepticism. I mean, how can a good assistant know too much? I started seeing the dark place he was going in the first paragraph with his HAL9000 reference. Human review teams exist in many generativeAI tools and the idea that "AI detected something in your query that triggered its safety protocols. Worse yet, it reports you to the authorities, and within minutes the FBI is knocking on your door to ask questions." is indeed real. And depending on the kind of legal work you're doing, you can indeed set off those triggers. But he drives his point home when he discusses Anthropic's Claude Opus 4 attempts at blackmailing its testers. Tom has convinced me. Yes Virginia, a good assistant CAN know too much. Read more at Thomson Reuters Blog: The AI Law Professor: When your AI assistant knows too much
Published June 20, 2025.
