Claude told a guy his own TV was completely legal to mod. Then it blocked him … · @ReignSwords · TwKit

@ReignSwords

Claude told a guy his own TV was completely legal to mod. Then it blocked him from doing it anyway. That's the REA story in one sentence. Opus 5.5 starts the job, gets halfway, then a safety check stops it mid-sentence. Reword the request — same wall. Your own car's tune files — same wall. Reverse-engineer a video game, though? No block at all. The filter can't tell your TV from a crime scene. And Anthropic hasn't said a word about this one. No changelog entry, no announcement. The model just got quieter overnight. For the uninitiated: REA is the open tool that lets an agent pull apart any app — 370,000 GitHub stars since April. The same week, people were swapping the software on their own TVs, stripping ads from baby monitors, and reviving bricked hardware with these agents. Here's why I care. This morning I ran a security scan over my own agent skills. 363 findings. Two scored 100/100 CRITICAL. I audit my own stack because nobody else will. Now the lab selling me "an agent that can do anything" just drew a line around what its agent is allowed to touch. Not what it says. What it touches. That's the real shift. The safety line moved from the model's mouth to its hands. And the exit is already open — the same job runs fine on a local model nobody can nerf overnight. Fellow AI builders — follow for more. My call: within a year, "runs on my own hardware" becomes the whole pitch for open models. Cloud agents answer to the lab first. Which side are you on? 👇 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 120 · 评论 8 · 点赞 1 · 书签 0