Basically: If you run OpenClaw connected to any meaningful system you are not fit to design, program or run any kind of software.

[art_codesmith@toot.cafe]

@tante Question: what if you name it?

[ftranschel@norden.social]

@tante The similarities to the concept of "unprotected intercourse" are more than striking here.

[tante@tldr.nettime.org]

@art_codesmith You should be forced into getting therapy

[toolsontech@pkm.social]

@tante If the people doing that would understand security they would be very upset.

I'm not worried about the competent people, I'm worried about the majority that doesn't even consider why giving access is a bad idea.

[art_codesmith@toot.cafe]

@tante So um... funny thing I read on LinkedIn today... #DHH

[tomw@mastodon.social]

@tante I agree BUT even if you run it on a Mac Mini or VPS or whatever without giving it your own data or accounts, it can still go and bother other internet users, for example by emailing them, sending spurious Github pull requests, making phone calls (!), the recent case of one making a hostile blog post about a named person, and so on.

Even if you make it not a danger to you, it remains a danger to others.

[art_codesmith@toot.cafe]

@tomw @tante The thing becomes then, why would you run it without giving it access to your accounts? The whole point is that it can Do Stuff For You.

[larsmb@mastodon.online]

@tomw @tante We desperately need a few incidents where their operators are identified and legally held responsible.

[tante@tldr.nettime.org]

@larsmb @tomw they are identified: By OpenAI's and Meta's and Anthropic's HR department to offer them jobs

[larsmb@mastodon.online]

@tante @tomw I didn't mean the one or two people who had them vibed, but the masses who run and operate them even more recklessly.

[lennybacon@infosec.exchange]

@tante Unfit… or “security expert at Meta”

https://www.golem.de/news/bei-sicherheitsexpertin-ki-agent-leert-unerwartet-ganzen-e-mail-posteingang-2602-205762.html

[vbabka@mastodon.social]

@tante more like OpenClown, huh.

[quietewe@urbanists.social]

@art_codesmith @tante lol he’s so proud of his agent, and then his story amounts to “I told the AI to do something. It couldn’t and didn’t have any idea how to proceed. So I told it to do a different thing first. And it did that, all on its own, no prompting from me! Then it did the first thing I told it to do. With no direction at all!”

Like awww it’s almost sweet that you don’t know what prompting and direction are, David! Truly groundbreaking.

[xoriff@hachyderm.io]

@tomw @tante we've gotta put some kind of regulation in place for this. What happens if an agent does something illegal? Presumably someone needs to be held accountable (I would hope it would be the operator). So there's got to be some way to link agent to operator. Otherwise, where's the accountability?

[lain_7@tldr.nettime.org]

@tante

“Meaningful system” here, includes a network.

[jukkan@mstdn.social]

@tante Here's the corporate vice president of Microsoft Word running #OpenClaw to organize his family tasks and comms: https://www.omarknows.ai/p/meet-lobster-my-personal-ai-assistant

Don't worry, I'm sure no malicious actors will ever target his setup to compromise #Microsoft network. I mean, they'll only have access to all info and location of his loved ones. Surely there's zero monetary value in that...

[jukkan@mstdn.social]

@tante OMFG. He has now connected #OpenClaw to his smart home AND published a full playbook of his setup:
https://lobster.shahine.com/

A senior leader at Microsoft basically gave the keys to his front door to autonomous GenAI agents.

[tante@tldr.nettime.org]

@jukkan professionals

[panda@pandas.social]

@lennybacon

"Das wird nicht nochmal passieren."

@tante

[patrick_h_lauke@mastodon.social]

@jukkan @tante fucking hell, some people REALLY crave having a personal assistant a la JARVIS. i particularly find the bit at the end creepy, where the AI compliments him on how "well written" his post is. validation/praise sycophant machine...