1Password for Claude Lets AI Log In Without Seeing Your Passwords

[FatLouie]

The best part is Claude will automatically transfer your bank account balance to my account.

Dave: "Claude...please return my money to my account."

Claude: "I'm sorry, Dave. I'm afraid I can't do that."

😁

 

[rembert]

I understand how this is working and I understand how 1Password believes this is secure as the AI tool won't get hold of the password. But, what if some AI prompt injection changes the behaviour of the AI tool and does use the 1Password function to view the password and record it... I have hacked many an AI system and support chat bots which didn't want to provide me certain answers but ultimately did by changing context and using circumventing prompts. Remember how Instagram got hacked not so long ago: the hackers just convinced the AI chatbot to add a new, attacker-controlled email address to the targeted account, allowing them to reset the password. It's often that easy.

 

[LordArchie]

That is exactly what this announcement is about avoiding - 1password handles it, so Claude never sees the credentials.

Problem is that the "headline readers" don't actually take the time to understand what 1PW and Claude are doing here.

 

[Kriss_De_Valnor]

The best part is Claude will automatically transfer your bank account balance to my account.

Do you mind if Claude transfers my mortgage balance to your account? 🤣

 

[Reason077]

Do people really trust these AI agents that much?

What could possibly go wrong? 🤔

 

[kevcube]

Can I be your AI agent, you can tell me everything. Starting with your SS and birthdate. No? But it's okay for data harvester to know to sell it to everyone around the world? weird priorities.

sure. I'll pay you $100 a month and you need to be on-call 24/7 and do inordinate amounts of work for me.

 

[Christopher Pelham]

It's all about "why would I want to do this?" The article talks about how secure and safe this is but not a single reason why it's worth doing. So it's only natural that folks will fill that gap in with their own assumptions. Personally when somebody goes out of their way to say how safe and secure something is without any concrete reasons why I should buy it, then I get suspicious which then quickly turns into mockery.

There are so many reaons this is worth doing. AI agents can redesign web sites, start and run companies, analyze and work with files in your Dropbox account, etc, almost anything you can think of that be accompished with a computer. To do many of those things, they need access. AI is so much more than a chatbot that answers your questions. Making access to login credentials more granular and protected makes all this safer.

 

[budselectjr]

Can I be your AI agent, you can tell me everything. Starting with your SS and birthdate. No? But it's okay for data harvester to know to sell it to everyone around the world? weird priorities.

Have you even used proper agents? You dont give them either pieces of that info and the permissions are so granular it can be locked down to not even have web access.

 

[123jmail]

There is nothing actually wrong with this.

If I want an AI agent to log into a website and do something for me, it needs the username and password. This way I get to see what it’s doing, and specifically approve that password only.

People get crazy with AI stories. The risk here is that the agent will do something unintended with the website once it’s logged in, not that it will steal all my passwords, because the way this is set up, it can’t.

I agree partly. I’m not worried about AI stealing the passwords, instead I’m not interested in this feature precisely because I don’t trust AI to do only what I want it to do.

I’ve read to many horror stories of AI taking unintended initiative and wrecking people who have way more tech sense than me. I’m not seeing enough benefit for the potential liability yet.

 

[thomasWmac]

sure bro, no credentials for AI. Let's trust the company that stole everything from the internet. Gosh how much more convenience do we need in our lives?

Since the passwords are handled by 1Password and not handed to Claude you do not have to trust Anthropic. If you use this you are required to trust 1Password. (And, of course, trust Claude to not do something bad with the access the password provided. Like, I don't know, changing the password...)

 

[Alfonson]

Paying for a password manager? HAHAHAHAH
AI looking at my passwords? HAHAHA what could go wrong! No security risk here!

Found the comment that reads more crazy than the article

 

[Xtrmist]

Can anyone say "Bad Idea"?

Bad Idea

 

[california_kid]

Problem is that the "headline readers" don't actually take the time to understand what 1PW and Claude are doing here.

Maybe or just as probable that they read the full article and decided it was all BS and lies. Everyone is entitled to their opinions but Google and Meta gave plenty of reasons to be skeptical about what a company claims and what it actually does in regards to security and privacy.