This article appears to be 100% AI. I guess there's some irony that a company ships an AI feature and someone else uses AI to come up with criticisms of that feature. But the article... doesn't actually say anything?
It's just full of weird, generic short-sentence LLMisms ("Detection is observation.", "Changing the password is authority.", "The security benefit is real.", "That is a meaningful improvement.", "This is not just text generation. It is an agent taking action with a sensitive credential.", ...). It doesn't offer any insights into the actual architecture that Apple came up with, whatever it might be. It doesn't propose a better design, other than a bunch of super-generic things that apply to every single software project ever ("The system should verify the exact website and account before filling or changing anything.", "This feature deserves focused adversarial testing during the beta period."). So... it's upvoted just because the title mentions Apple and AI?
Back in 1984, I wrote the original "obvious password detector".[1]
It just checks whether a password has English language trigram stats. This prevents dictionary attacks.
There's this standard that is being worked on by the people working on the Passwords app at Apple (They are active on Mastodon, and often talking about that) which will probably be helpful for this feature too: https://www.w3.org/TR/change-password-url/
Very curious if they're implementing browser driving themselves or using an off-the-shelf library like stagehand, browser-use, etc. to drive the DOM. Hopefully they open source it if it's in Swift.
A11y-tree alone is not enough for many sites because lots of auth stuff happens in OOPIF frames that need special handling/stitching/interactive element filtering.
There's also the issues of many captchas around auth stuff being implemented using canvas elements (that are hard to instrument for browser agents without relying on CUA). Can their on-device 3B model really handle accurate CUA driving? I guess we'll see...
I wonder whether the AI generated password that you allow to be created on your iPhone in the Passwords app can be recovered and added to whatever password manager you might be using on Windows or Linux desktop.
It seems like this is a great way to lock oneself out of access to an account on some of the devices that they own that do not have access to the Passwords data storage.
I can see where this can be a benefit in helping users secure their accounts with stronger passwords but I think that there is a lot of potential for this to become a real problem.
This could have nuclear level consequences. Imagine somehow your keychain is compromised. Using a change password URL means an attacker could literally lock you out of every account at the same time
this only really changes things for obscure sites. there's already automation readily available for all the popular social media, banks, crypto sites, etc.
This is a great article except the "That can happen for plenty of boring reasons" list. Almost each of those reasons is completely unrelated to AI and can happen even if you attempt the change 100% manually with or without a password manager.
Yes, also immediately thought of all the endless ways this could go wrong and end with someone losing access to their account, which depending on their account could be trivial or life altering, especially if their loss ends up being someone else's gain. Apple takes baby steps so I'm sure this will be limited in nature and most likely will get delayed until fully tested, but I'd definitely avoid testing during betas (with any real accounts that is).
This is not that. This is an agent autonomously navigating through the website to change your password. You don’t even need to be navigated to the site. The app gives you a list of sites with insecure passwords and you just hit a button labeled “Change” or something like that and it kicks off a process to change it. If there’s was a standard API for changing passwords, then sure. But this is far beyond just suggesting a strong password and offering to store it for you when you are navigating the site yourself. And given how often THAT simple functionality goes wrong, in my experience, I don’t have a lot of hope that this will work out well. To be clear, that’s not Apple’s fault. Some sites have garbage authentication and password change workflows.
I can remember two passwords: the one that gets me into my laptop, and the one that gets me into my password manager. And this feature requires one to use Apple's default password manager, ergo...
And I shouldn't remember the first one, I just haven't gotten 'round to setting up the Yubikey on the laptop just yet.
As per the demo, in order for Siri to rotate your passwords "for you", you have to open the Password app, go to their dashboard on weak or exposed passwords, and click a button asking it to rotate your password account by account.
So yes. It's off by default. You have to affirmatively use the feature. (This is purely based on what I remember from the demo, mind you. I have not used the feature.)
This one is getting a lot of undue flak. Not only does it require explicit confirmation, it’s also contained entirely within the passwords app which already has access to all your passwords because you chose to trust it.
If you use this app, open it and look at how many entries fall under the “security” section. Everyday another password is compromised and added to the list, just too many to keep up. So, albeit apprehensively, I for one appreciate this feature.
I already let 1Password generate all my passwords, so as long as they're just invoking tools with AI rather than having it attempt manually, it doesn't seem like such a big deal?
If they already use devices outside the Apple ecosystem, they're not using the Passwords app, or they're using the plugins that get you access to it in other ecosystems.
It's just full of weird, generic short-sentence LLMisms ("Detection is observation.", "Changing the password is authority.", "The security benefit is real.", "That is a meaningful improvement.", "This is not just text generation. It is an agent taking action with a sensitive credential.", ...). It doesn't offer any insights into the actual architecture that Apple came up with, whatever it might be. It doesn't propose a better design, other than a bunch of super-generic things that apply to every single software project ever ("The system should verify the exact website and account before filling or changing anything.", "This feature deserves focused adversarial testing during the beta period."). So... it's upvoted just because the title mentions Apple and AI?
Everything is so much more complicated now.
[1] https://www.animats.com/source/obvious/obvious.c
For anybody else trying to know what else the .well-known URI can hold: https://en.wikipedia.org/wiki/Well-known_URI#List_of_well-kn...
I'd have really preferred another term: registered, reserved, defined, meta -- or really anything else.
At any rate this is just the first step towards a first-party agentic OS.
I should also add this is only if you have iOS 26 or newer.
A11y-tree alone is not enough for many sites because lots of auth stuff happens in OOPIF frames that need special handling/stitching/interactive element filtering.
There's also the issues of many captchas around auth stuff being implemented using canvas elements (that are hard to instrument for browser agents without relying on CUA). Can their on-device 3B model really handle accurate CUA driving? I guess we'll see...
It seems like this is a great way to lock oneself out of access to an account on some of the devices that they own that do not have access to the Passwords data storage.
I can see where this can be a benefit in helping users secure their accounts with stronger passwords but I think that there is a lot of potential for this to become a real problem.
And I shouldn't remember the first one, I just haven't gotten 'round to setting up the Yubikey on the laptop just yet.
this also requires the passwords app to even function. so this should be a non-issue.
So yes. It's off by default. You have to affirmatively use the feature. (This is purely based on what I remember from the demo, mind you. I have not used the feature.)
If you use this app, open it and look at how many entries fall under the “security” section. Everyday another password is compromised and added to the list, just too many to keep up. So, albeit apprehensively, I for one appreciate this feature.
A good chunk of people do use devices other than apple eco system one's and if they try to login and then suddenly, you can't!