This blog is on the malwarebytes website. Mslwarebytes says in thr post thst its not fair to call this spyware. This was brought up on the windows side as well.
What is really going on: claude desktop is installing the hooks for the claude browser extension. If you install the browser extension, claude desktop can control the browser. This is the intended behavior so you can have an agent do something like “in the morning, access these three sites, pull down the data and create a newsletter for me” or “please check flight costs throughout the day on these sites” or whatever you want to access the browser for.
This is the whole reason you install claude desktop, to automate your computer.
Even if this was an opt-in feature the implementation is still terrible and a massive security hole. If id wanted the desktop app entirely and solely for this purpose i still would not expect my browser extension to have full access to my computer. I understand the app does, not the browser extension.
No matter how you twist and turn this situation Anthtopic has still introduced a major security issue in their application. It might be a bit far to call it malware immediately but it sure does open up a massive attack vector to take advantage of.
The fact that the end user is not even informed or have a choice about this makes it all the more problematic and Anthropic not commenting on it makes me think its either intentional or at the very least already known.
Your comment seems very dismissive in the way you phrase this as intended behaviour. A security flaw like this can impossibly be intended behaviour.
In my previous comment i also say thats calling it malware is a bit far-fetched but the security issues are absolutely there and should not be dismissed as “intended behaviour”. Especially not by a company like Anthropic.
I am not well versed in extension development but is there anything stopping me from making an open source extension and just defining the ID as one of the three in the article? It most likely couldnt be released via the chrome addon store but if it is installed outside of thar? And how are these IDs read after install, could it potentially be altered by something from the outside?
I immediately see so many flaws with this implementation it is worrying that a company the size of Anthropic does this.
The article says that is the intended use, I agree this is just bad implementation, but it’s bad because it not only allows control one way, from the app to the browser, it also allows it the other way: browser extensions with an ID that matches one of the allowed ones can access userspace, without asking. That is a huge attack surface that is installed without any consent.
This is a little disengenuous…the browser extension ≠ the desktop app. Some people install the app and only use the chat feature. Some use cowork but would never want to use the browser extension. Assuming that installing a desktop app means you should also want the browser extension is just bad logic.
You cant access the browser unless you insta the extension. The desktop app just places jooks for the extension if it is ever installed. It wont work with out the extension
I agree that this doesn’t rise to the level of “spyware,” but it is extra sneaky/slimy, and it absolutely, IMO, makes your system less secure for no good reason. They could just have a prompt in the UI the first time you attempt to use a feature that requires the native messaging host, which says something like “we need to install extra software to communicate with Chrome, OK?” This is the ethical thing to do.
It’s especially sketchy that they’re preemptively installing it in the right directories for multiple Chromium-based browsers, even ones that aren’t installed on your system.
Its not sketchy just lazy. One observation i have made eith anthropic is that they are great at amking a model but louzy at app development. There apps tend to have that “scientist learned python to help them at work” vibe. Which is always a security nightmare.
I disagree, it’s definitely sketchy. Going out of your way to install the messaging host for a half dozen different Chromium forks is going out of your way do something behind the user’s back; it’s the opposite of lazy.
I imagine it’s more of a vibe-coded “make sure the end users have all the files they’ll need to be ready to go” prompt, and it’s Claude that “decided” to just have all the files from the get-go
Is that not what we were doing? I’m not disagreeing that it’s scummy that they’re installing unnecessary files, just speculating that’s it’s ineptitude rather than malicious. Hanlon’s Razor and all that. Considering the downvotes on my comment I may have misread your comment.
It also uses your credentials to do so and doesn’t ask any permissions for any of it including whatever else it wants to do outside the browser sandbox where it lives. Anthropic can easily remedy the situation but they didn’t set it up that way. And the question is why.
Not calling it spyware is like not calling McDonald’s “food”. While technically true, it’s just how it works.
I don’t think it’s actually doing anything nefarious yet. fwiw.
This blog is on the malwarebytes website. Mslwarebytes says in thr post thst its not fair to call this spyware. This was brought up on the windows side as well.
What is really going on: claude desktop is installing the hooks for the claude browser extension. If you install the browser extension, claude desktop can control the browser. This is the intended behavior so you can have an agent do something like “in the morning, access these three sites, pull down the data and create a newsletter for me” or “please check flight costs throughout the day on these sites” or whatever you want to access the browser for.
This is the whole reason you install claude desktop, to automate your computer.
Even if this was an opt-in feature the implementation is still terrible and a massive security hole. If id wanted the desktop app entirely and solely for this purpose i still would not expect my browser extension to have full access to my computer. I understand the app does, not the browser extension.
No matter how you twist and turn this situation Anthtopic has still introduced a major security issue in their application. It might be a bit far to call it malware immediately but it sure does open up a massive attack vector to take advantage of.
The fact that the end user is not even informed or have a choice about this makes it all the more problematic and Anthropic not commenting on it makes me think its either intentional or at the very least already known.
The security issue, as the blog says, is that it trusts any extension with the id. So if you can spoof the extension you have access.
What i was saying is that its not spyware. Which is a different issue.
Your comment seems very dismissive in the way you phrase this as intended behaviour. A security flaw like this can impossibly be intended behaviour.
In my previous comment i also say thats calling it malware is a bit far-fetched but the security issues are absolutely there and should not be dismissed as “intended behaviour”. Especially not by a company like Anthropic.
I am not well versed in extension development but is there anything stopping me from making an open source extension and just defining the ID as one of the three in the article? It most likely couldnt be released via the chrome addon store but if it is installed outside of thar? And how are these IDs read after install, could it potentially be altered by something from the outside?
I immediately see so many flaws with this implementation it is worrying that a company the size of Anthropic does this.
There are many flaws. I am in no way cintradicting it or dismissing it.
Now you’re just making stuff up.
Excuse me if im misunderstanding something but what exactly am i making up?
The article says that is the intended use, I agree this is just bad implementation, but it’s bad because it not only allows control one way, from the app to the browser, it also allows it the other way: browser extensions with an ID that matches one of the allowed ones can access userspace, without asking. That is a huge attack surface that is installed without any consent.
This is a little disengenuous…the browser extension ≠ the desktop app. Some people install the app and only use the chat feature. Some use cowork but would never want to use the browser extension. Assuming that installing a desktop app means you should also want the browser extension is just bad logic.
You cant access the browser unless you insta the extension. The desktop app just places jooks for the extension if it is ever installed. It wont work with out the extension
I agree that this doesn’t rise to the level of “spyware,” but it is extra sneaky/slimy, and it absolutely, IMO, makes your system less secure for no good reason. They could just have a prompt in the UI the first time you attempt to use a feature that requires the native messaging host, which says something like “we need to install extra software to communicate with Chrome, OK?” This is the ethical thing to do.
It’s especially sketchy that they’re preemptively installing it in the right directories for multiple Chromium-based browsers, even ones that aren’t installed on your system.
Its not sketchy just lazy. One observation i have made eith anthropic is that they are great at amking a model but louzy at app development. There apps tend to have that “scientist learned python to help them at work” vibe. Which is always a security nightmare.
I disagree, it’s definitely sketchy. Going out of your way to install the messaging host for a half dozen different Chromium forks is going out of your way do something behind the user’s back; it’s the opposite of lazy.
i think a lot of their stuff is ai coded.
That’s not really relevant to the point I’m making.
I imagine it’s more of a vibe-coded “make sure the end users have all the files they’ll need to be ready to go” prompt, and it’s Claude that “decided” to just have all the files from the get-go
Ah yes let’s wildly speculate about how they may or may not be writing software.
Is that not what we were doing? I’m not disagreeing that it’s scummy that they’re installing unnecessary files, just speculating that’s it’s ineptitude rather than malicious. Hanlon’s Razor and all that. Considering the downvotes on my comment I may have misread your comment.
I think that’s too generous. As if they didn’t think about this.
I would not assume a chatbot app would auto create hooks into a browser like this. That’s not a reasonable assumption to make.
It also uses your credentials to do so and doesn’t ask any permissions for any of it including whatever else it wants to do outside the browser sandbox where it lives. Anthropic can easily remedy the situation but they didn’t set it up that way. And the question is why.
Not calling it spyware is like not calling McDonald’s “food”. While technically true, it’s just how it works.
I don’t think it’s actually doing anything nefarious yet. fwiw.
Side question, are the typos intentional?
Mobile keyboard without spellcheck, I make thr exact same typos as thst poster with my thick fingers.
Or it’s an iPhone, they sometimes don’t input pressed letters or input the wrong one.