OpenClaw Browser Relay

OpenClaw introduced a Chrome extension called Browser Relay giving OpenClaw agent(s) control to existing user controlled browser tab.  One can also develop custom agents, independent of OpenClaw, to directly control opened browser tabs via Chrome DevTools MCP (the underlying tech Browser Relay is built on). 

I am curious what everyone thinks of these new AI agent capabilities, how they will affect Torn City players and whether Ched and the dev team have adequate tools/instrumentations to catch script abuse as given the push to give AI agents more direct control to our browsers and apps.

Well this can't be good for browser-based games. Do AI agents have janky mouse movements or is it pixel perfect?

I think a fed in this game is not the first thing you should worry about when you give the clankers full control over your browser. 😂

Load image

People dumb enough to use openclaw and give it full access are the people dumb enough to think they can get away with it in torn and be stupidly wrong. So torn is safe.

Sounds like circular reasoning.  It's entirely feasible to setup a docker container with VNC + Chrome browser + OpenClaw, sign-in to Torn (and only Torn) and let AI do its thing.  This way, AI is contained only in the little sandbox and cannot do anything to harm the host system.

I think this is the case yeah. If there's a "human error" variable with OpenClaw, then you last a few extra months, but no one is naturally able to perfectly minmax their account long term.

I'd expect to see a small string of feds over it but nothing major.

I think when AI browsers first started being talked about, there was a news article somewhere about a guy using one. The browser went completely rogue when he asked it to do something and nearly bought something online despite that action having precisely no connection with what the reporter asked it to do.

That's just one of the many risks of using AI browsers. 

Full control of headful (as opposed to headless) browser in a container is generally considered safe if you only give it limited login credentials.

[image: media1.tenor.com]

How about 2FA? Still safe?

One can simulate mouse movements and click events and randomize the coordinates.  I have made a POC early last year to take a screen capture of a web screen (NOT TORN), pass it on to Gemini and let it tell me a specific button's on-screen coordinates + size and then use JS to generate mouse click event with some randomness.  It's also fairly easy to simulate human like mouse movements.

https://www.theguardian.com/technology/ng-interactive/2026/mar/12/lab-test-mounting-concern-over-rogue-ai-agents-artificial-intelligence

https://trufflesecurity.com/blog/claude-tried-to-hack-30-companies-nobody-asked-it-to

Safe my ass lol

Edit: And those are literally just two things I randomly read about last week. 

Lol, JS...

If I understand your question correctly, 2FA is only used for login, after login, it doesn't play a role until next login.  In a docker container, you don't need to give AI the ability to login, instead, you login first, "attach" the AI and give it a task so it doesn't know your login.  

And there is:

We Let AI Run Our Office Vending Machine. It Lost Hundreds of Dollars

Yeah, so the AI with control over your browser (and therefore at least your account - do you have a payment method saved for Torn by any chance? ;) couldn't only go rogue by itself, it ALSO could be manipulated by anyone in Torn. Great point. 😂

Or Python using PyAutoGUI which can control apps beyond what's in the browser.

[image: media1.tenor.com]

I wouldn't trust it with any of my passwords or payment info and I will setup a strict sandbox environment/container in a separate subnet and firewall routing rules.  

it ALSO could be manipulated by anyone in Torn

Yes but it could also be used for very specific and limited tasks, for example, monitor and auto update company inventory or help identify buy mug targets.