Can Invisible Watermarks Stop AI Junk? I'm Skeptical

Just a heads up. I saw that Nature report—Anthropic added invisible watermarks to Claude's outputs, tagging both text and images. They say it's to comply with the EU AI Act and also to curb "AI slop." I just started trying out Claude Code in my terminal these past couple of days, and right on cue, they've put a leash on it. Feels a bit weird.