RE: 7 Day Call for Consensus March 17, 2016 ARIA Working Group Resolutions

The goal of the password role is not to improve security. The goal is to improve user experience. It is no different than any other aria role in that. The web developer has already made a fake control with scripted behavior. The role just lets him tell the accessibility api what the control is supposed to be, so the accessibility api can tell the AT, and the AT can tell the user.

The password role does not prevent accessing the content of the password field from script. It could, on some platforms, prevent accessing from the accessibility API. However, if malware is using the accessibility api from the OS side, the machine is already compromised.

The attacks below rely on the bad actor being physically present to read the password over the user’s shoulder. The requirement for physical presence significantly reduces the risk.

From: Matt King [mailto:a11ythinker@gmail.com]
Sent: Tuesday, March 29, 2016 8:42 AM
To: 'Rich Schwerdtfeger' <richschwer@gmail.com>; 'John Foliot' <john.foliot@deque.com>
Cc: 'Léonie Watson' <tink@tink.uk>; 'ARIA Working Group' <public-aria-admin@w3.org>
Subject: RE: 7 Day Call for Consensus March 17, 2016 ARIA Working Group Resolutions

Since the discussions at CSUN, I have been giving this thought.

As I understand it, we have 2 types of security exposure:

1.       With the password role as defined: blind users could type in an ARIA password field where, because input is audibly obscured, they will be unaware if the input is visible, especially if it were intentionally visible for nefarious reasons.

2.       Without the password role: blind users could type in a field and expose their password audibly because the screen reader didn’t automatically override its key echo function that announces which keys are pressed on the keyboard.

Note that without the password role, the user is in control. All screen readers have a key for turning off key echo. So, no password role is the safer course if we can not resolve the conflict.

However, in the interest of preventing exposures for screen reader users who may have forgotten which key turns off key echo, and who do not have head phones handy, and who do not have time to lookup the required information, I have given some thought to requirements we could reasonably place on support for the password role that would resolve the conflict.

We can not place a requirement on browsers to ensure the input is visually obscured. That would be requiring user agents to change behavior for all users, which would violate our long-standing principle to not do that.

Instead, could we place a normative requirement on screen readers to do 2 things when a password field is encountered:

1.       Automatically turn off key echo while focus is in the password field.

2.       If key echo was on, echo exactly what is typed in the field instead.

Matt King

From: Rich Schwerdtfeger [mailto:richschwer@gmail.com]
Sent: Monday, March 28, 2016 3:25 PM
To: John Foliot <john.foliot@deque.com<mailto:john.foliot@deque.com>>
Cc: Léonie Watson <tink@tink.uk<mailto:tink@tink.uk>>; ARIA Working Group <public-aria-admin@w3.org<mailto:public-aria-admin@w3.org>>
Subject: Re: 7 Day Call for Consensus March 17, 2016 ARIA Working Group Resolutions

John,

First, authors are already creating these custom password fields to say that they must behave exactly like an HTML5 password field does not make any sense. The Microsoft security people stated that people were doing this already and adding a password role does absolutely nothing to stop them from doing so.

The far bigger security issue is that despite the fact that they are creating one of these custom fields we give the author zero vehicle to tell the screen reader to not ECHO the keyboard keys being type for all to hear in the room. They don’t echo the keys you actually see which are obscured. they echo the text typed coming in from the keyboard.

What is your solution to prevent this? Yelling at them to use HTML5 passwords is like our shouting at the moon.

I want to see a real solution here.

Where are we asking a browser to change their UI? Browser vendors have been very clear to tell us that we cannot require them to change their UI based on ARIA. On this I see no win although I agree with you that it would be could here.

So, the net of this if we don’t include the role we continue to leave users exposed with a security hole where everyone can hear the password they are typing unless they happen to have a headset on. Is that what you both want?


Rich Schwerdtfeger



On Mar 28, 2016, at 5:02 PM, John Foliot <john.foliot@deque.com<mailto:john.foliot@deque.com>> wrote:

Hi Rich,

After chatting with some folks at CSUN, I share Leonie’s concerns. Unless all of the browser vendors and screen readers are going to programmatically treat the role=”password” *EXACTLY* like input type=”password” I too see a serious security/privacy concern.

For example, what should we expect with this piece of code: <input type=”text” role=”password”>?

Will screen readers announce “star, star, star” while displaying “Secret PIN #” in the text field, in the clear and open?  (Saying they shouldn’t do that  isn’t enough, I just did it and so others will as well)

Likewise for a scripted input, perhaps something like <div class=”Input_Field” role=”password”>: how do we guarantee end users that the scripted input *is* being treated like an actual password input, and isn’t a fishing spoof on non-sighted users? Companies like IBM would likely never do that, but IBM isn’t the only folks writing code out there :D

I also understand that this is needed for SVG, so my concern is not that we need a “something”, but rather, again, we’re asking browser vendors to change their UI based upon an ARIA attribute, something that they have refused to do in the past, as for example here: https://lists.w3.org/Archives/Public/public-pfwg/2015Sep/0172.html


JF

From: Rich Schwerdtfeger [mailto:richschwer@gmail.com]
Sent: Monday, March 28, 2016 5:37 PM
To: Léonie Watson <tink@tink.uk<mailto:tink@tink.uk>>
Cc: ARIA Working Group <public-aria-admin@w3.org<mailto:public-aria-admin@w3.org>>
Subject: Fwd: 7 Day Call for Consensus March 17, 2016 ARIA Working Group Resolutions

Leonie,

Did my response address your concern? Microsoft confirmed that people were creating their own custom passwords in the wild and there is no ARIA role to indicate to the AT that this is a password and to tell the AT to NOT echo the password text as you type it. This would facilitate that.

Rich


Rich Schwerdtfeger




Begin forwarded message:

From: Rich Schwerdtfeger <richschwer@gmail.com<mailto:richschwer@gmail.com>>
Subject: Re: 7 Day Call for Consensus March 17, 2016 ARIA Working Group Resolutions
Date: March 20, 2016 at 3:59:23 PM CDT
To: tink@tink.uk<mailto:tink@tink.uk>
Cc: ARIA Working Group <public-aria-admin@w3.org<mailto:public-aria-admin@w3.org>>

Leonie,

On the other hand, a screen reader could announce the characters being typed and not know to not do that. Furthermore, people are creating these things today and there is no way to know that the textfield is a password field. Would you prefer to not know?

I don’t understand how your statement supports your argument. Incidentally,we did vet this with the Microsoft browser security people before agreeing to add it to the spec. Microsoft stated that people were creating their own password textbooks in the wild and there is no way for you to know that is what the textfield is.

Rich

Rich Schwerdtfeger



On Mar 17, 2016, at 3:06 PM, Léonie Watson <tink@tink.uk<mailto:tink@tink.uk>> wrote:

From: Rich Schwerdtfeger [mailto:richschwer@gmail.com]
Sent: 17 March 2016 19:12
To: ARIA Working Group <public-aria-admin@w3.org<mailto:public-aria-admin@w3.org>>
Subject: 7 Day Call for Consensus March 17, 2016 ARIA Working Group Resolutions

This is a Call for Consensus (CfC) to the Accessible Rich Internet Applications (ARIA) Working Group on the following resolution:

1. Accept Joanie’s addition of a new password addressing Action 2004:

https://rawgit.com/w3c/aria/password-role/aria/aria.html#password


I object to the password role. Unless I’m missing something, it leaves open the possibility that an AT will behave as though the characters input into the field are obscured, when visually they may not be. A screen reader user cannot be certain that their password is adequately protected from being observed.


Léonie.

--
@LeonieWatson tink.uk<http://tink.uk/> Carpe diem.