Claude Code can now take over your computer to complete tasks

But Anthropic urges caution as "research preview" safeguards "aren't absolute."

See full article...

 

[Niles Gazic]

You know what else Claude Code can do? It can f*ck right off. That's what it can do.

F*ck this AI bullshit. Who asked for this stupid future?

You're welcome.

Quoting you, since I can't upvote you further.

 

[Dr Spiff]

You know what else Claude Code can do? It can f*ck right off. That's what it can do.

F*ck this AI bullshit. Who asked for this stupid future?

You're welcome.

I feel the same, but still have days where I question if it’s just me not understanding. What if I’m wrong?

There’s not a single day going by without someone in my office suggesting how we use AI for this, that or the other thing.

Some stuff is about summarising information, which can be useful and sometimes even reasonably accurate.

Then there’s other stuff like my manager asking me to “feed the data into the AI” to see if it can “generate the same analysis code” as I’ve written. And when I ask what he thinks “the code” is, he doesn’t know. And when I ask how we can know if “the AI” understands how the data has changed over time and how to account for that, he doesn’t know. But he’s really enthusiastic about us all using “AI” a lot more.

I’ve been researching off grid living a lot recently..

 

[launcap]

You know why. It begins with “L” and ends with “Aziness”.

AKA "I want something for nothing"..

 

[Space Filling]

the model is trained to avoid “risky operations” such as moving or investing money, modifying files, scraping facial images, or inputting “sensitive data.” But the company also warns that such training safeguards “aren’t perfect” and “aren’t absolute,” meaning that “Claude may occasionally act outside these boundaries.”

I wouldn't call those operations "risky", I'd call them illegal.

Not only that, but I don't want "safeguards" that aren't because Claude can act outside those boundaries. And I don't want them to be "trained", I want them to be absolute. Why the fuck would Claude ever consider scraping facial images (or any other images)? Why can't Claude "know" that private data must never be accessed, let alone "scraped".

Because what that disclaimer is really saying is: "Claude may occasionally move and invest money, modify personal files, scrape images, and access sensitive data, all without asking you or informing you."

 

[Eldorito]

On the one hand, this is stupid and I will never touch it.

On the other, I get a daily report from a company that involves following a link, clicking through two other pages and clicking some random JS download button to download the file. Any kind of automation tool always fails at it. I really want a tool that can click the totally obvious things on the page to save me 2 minutes a day. So I kinda hope this works eventually.

 

[MilanKraft]

I think we do, because I don't think the worst thing you can imagine is the actual worst thing that could happen. Sadly.

In fact, I'm willing to bet there will be a story in a month or two (being hopeful it's not sooner) about some disaster relating to this, where people are shocked (shocked!) that something horrific happened.

100%.

Not a fan of cliche over-use at Ars (or anywhere else), but "What Could Go Wrong" is the question every single person should ask, every single time something like this is floated to them as even a possibility, let alone a suggestion for them personally. Fallout from this type of tech (including clawbot horrors) is likely to get much MUCH worse than anything we've seen with malware that takes hold via corporate or personal phishing expeditions. This is my favorite quote...

the model is trained to avoid “risky operations” such as moving or investing money, modifying files, scraping facial images, or inputting “sensitive data.” But the company also warns that such training safeguards “aren’t perfect” and “aren’t absolute,” meaning that “Claude may occasionally act outside these boundaries.

...right so, "even though we told it not to do these dangerous things, sometimes it's going to do those dangerous things anyway, but we thought 'what the heck!' we'll let you guys try it out anyway!" As they say on the internets: "telling us your product is mostly unreliable and completely unpredictable, without actually telling us..."

Anthropic my ass. I know misanthropic when I see it (I'm one of the originals)... even if their stated intent is otherwise. PR doesn't matter; what you put out into the world to wreak havoc is what matters, you "move fast and break things" assholes. And if, like your mass copyright theft which you are now settling a massive class action suit for, it all goes very wrong —imagine someone lets this wreck-it-tech loose in a hospital or financial trading environment — I hope your company gets curb-stomped out of existence in court, and the founders who green-lit a mostly unvetted product for public use, jailed and convicted... even if it means country club prison in the end (which it surely would).

 

[noteonleaf]

This agentic apps only useful as bot for automated playing any RPG and MMORPG. So I'm not wasting time grinding by myself .

 

[BroastedToaster]

I look forward to being locked out of my computer while watching this delete the win32 virus for me.

 

[graylshaped]

I feel the same, but still have days where I question if it’s just me not understanding. What if I’m wrong?

There’s not a single day going by without someone in my office suggesting how we use AI for this, that or the other thing.

Some stuff is about summarising information, which can be useful and sometimes even reasonably accurate.

Then there’s other stuff like my manager asking me to “feed the data into the AI” to see if it can “generate the same analysis code” as I’ve written. And when I ask what he thinks “the code” is, he doesn’t know. And when I ask how we can know if “the AI” understands how the data has changed over time and how to account for that, he doesn’t know. But he’s really enthusiastic about us all using “AI” a lot more.

I’ve been researching off grid living a lot recently..

Leave spellcheck on and tell him all your documents are reviewed by AI.

 

[Anton Longshot]

"Complete tasks"?
You mean clicking the "x", right?
Wow amazing I'm so impressed I have no words.

 

[DKlimax]

No Linux desktop UI client?! Seriously?!

Be glad or you'd get acquainted with most expensive rm -rf / stories...

On the one hand, this is stupid and I will never touch it.

On the other, I get a daily report from a company that involves following a link, clicking through two other pages and clicking some random JS download button to download the file. Any kind of automation tool always fails at it. I really want a tool that can click the totally obvious things on the page to save me 2 minutes a day. So I kinda hope this works eventually.

Wouldn't script based on Webdriver work?

Just waiting for the windows release to be jailbroken and have it delete system32.

What jailbreak? What you mean by it in this context? In any case, Administrators do have sufficient rights to System32, but there are multiple safeguard to prevent accidents. But I have to admit beside hosts I never needed to delete or replace file manually there.

I think we do, because I don't think the worst thing you can imagine is the actual worst thing that could happen. Sadly.

In fact, I'm willing to bet there will be a story in a month or two (being hopeful it's not sooner) about some disaster relating to this, where people are shocked (shocked!) that something horrific happened.

Two weeks.

View: https://www.youtube.com/watch?v=-KFH--_cdiI