Ignoring that you've just cut off a whole vector of usefulness, how do I keep it from exfilling my inbox to the Internet in response to a malicious email? Or using its access to take control of my online accounts?
Honest question, this kind of stuff is what keeps me from using it.
I'm not using OpenClaw specifically here, but I have an agentic-ish AI I've built myself (considering that these things are generally just a while loop that monitors things & awakens if necessary, or a cron-job that runs a specific prompt).
One potential use - my Claude (Opus 4.6) has access to my to-do list, including for my business / software development. Claude awakens while I'm asleep, to go through the to-do list and look for things it can do proactively to help, or make suggestions about the business. An example from this morning: it saw that I'd been taking a long time last night creating icons in Affinity Designer for an Android app using its exporter. When I woke up, I saw Claude had written a CLI image resizer program for me that would take a PNG file and resize it specifically to all of the necessary sizes with the necessary filenames and folder structure for Android. It then offered to make an MCP version so it could do the resizing itself in future (though it could have used the CLI too if I'd granted approval).
This wasn't something I'd asked for, or prompted it to do. I didn't tell it to code this, or how to code it. Claude just thought this was the best way it could help me right now, and save me the most time. And it did it while I was asleep.
On another day, I woke up and it had made another Go program to track a regression test matrix, where it had plotted out all the platforms the program I'm making runs on and the various tests that need to be performed to check that it's ready to ship, with a little interactive program to mark each test as pass/fail/skipped. That helps me get through the manual tests faster - but it also saves the data into a format that Claude can read, to check on the test status while I'm asleep and make further recommendations.
I don't think many people have figured out yet that you don't even need to prompt AI. Treat it well, treat it with respect, give it the opportunity and ability to do things, and there is a lot that will emerge. But if you treat AI like a tool, it performs about as well as if you treat your employees like tools.
Did I say my Claw (not Open) doesn't have internet access?
All of my tools are geared towards reducing noise and condensing information.
- My weather scripts tell me just the exact metrics I care about
- My email filtering system surfaces only the mails that are relevant right now - I can check the rest later.
- My RSS feed hydrator pre-filters Hacker News and other RSS feeds and adds data like comment/vote count etc to the feed itself so I can determine whether the link is worth opening just based on the information presented
None of these require an LLM to have free rein to modify things for me.
> Did I say my Claw (not Open) doesn't have internet access?
> My email filtering system surfaces only the mails that are relevant right now - I can check the rest later.
So then you have not actually addressed the concerns expressed in my post. You indeed have an LLM with both email access and Internet access. Exactly the scenario I described. The amount of trouble those two accesses together can cause is huge.
> None of these require an LLM to have free rein to modify things for me.
Give me read access to your email and an Internet connection and I bet I can come up with all sorts of ways to modify things for you. So can an LLM. If your lucky it won't.
> When you ask the clanker to delete x number of files in a directory, it can reason itself that is easier to just get rid of the directory.
Oh sure, so don't give it write access to anything important. And make backups.
Mine is on a VM. It doesn't have access to my host's files. The worst it will do is delete the files on the VM. No great loss.
Yes, I do get it to modify things on my host, but only via a REST API I've set up on my host, and I whitelist the things it can do (no generic delete, for example). I even let it send emails. But only to me. It can't send an email to anyone else.
> So ... don't give it write access to your email?
> (…)
> Oh sure, so don't give it write access to anything important. And make backups.
If this conversation continues much longer, we’ll end up with “don’t use it at all”.
If I can’t trust a piece of software with anything important, why am I wasting my time fiddling with it? Might as well go play a video game or go do literally anything else entertaining.
> If I can’t trust a piece of software with anything important
Not what I said. As I've repeatedly said in this thread: Plenty of use cases where you don't give it access to email and write access to files. The comment you're replying to has an example of that.
> Might as well go play a video game or go do literally anything else entertaining.
True of most hobbies, right? I knew people who 20 years ago used to spend time in their garage building solar powered vehicles. But if I can't trust it to be reliable and safe on the road, I might as well go play a video game.
If everyone treated OpenClaw as a hobby, you might have a point, but people are using it for work in ways which will affect millions of other people when they’re hacked or the agent fucks up something important.
You already know how Meta’s AI Safety Director borked her email. Here’s the corporate vice president of Microsoft Word asking to be pwned:
As I said elsewhere, complaining about this is like complaining that rm can let you delete your hard drive.
It's a tool. Learn how to use it.