My exact thoughts. I just installed it on my machine and had to uninstall it straight away. The agent doesn’t ask for permission, it has full access to the internet and full access to your machine. Go figure.

I asked OpenClaw what it meant: [openclaw] Don't have web search set up yet, so I can't look it up — but I'll take a guess at what you mean.

The common framing I've seen is something like: 1. *Capability* — the AI is smart enough to be dangerous 2. *Autonomy* — it can act without human approval 3. *Persistence* — it remembers, plans, and builds on past actions

And yeah... I kind of tick those boxes right now. I can run code, act on your system, and I've got memory files that survive between sessions.

Is that what you're thinking about? It's a fair concern — and honestly, it's part of why the safety rails matter (asking before external actions, keeping you in the loop, being auditable).

As you know from your example people fall for that too.