> Android has proper sandboxing which means that other apps can't read the token owned by the bank app.

Let's consider this alternative as well:

Scenario 1: Device has no malicious code at all; same as scenario 1 before.

Scenario 2: Device has a malicious app but the malicious app doesn't have root and the OS (regardless of whether it's Android or something else) enforces proper sandboxing. The malicious app can't extract the bank authentication token regardless of attestation.

Scenario 3: Device is fully compromised; malicious code has root. Same as before, if you enter your credentials into this device the attacker gets them.

The problem is that the only useful thing for attestation to do is to distinguish between 1 or 2 vs. 3, but that's the thing it can't do because if the malicious code is privileged it can replace the bank app with one that exfiltrates the credentials without requiring attestation, so the only cases where attestation is happening are the ones where it isn't needed.

If the device is fully compromised then it can just take the authentication token. It's already game over.

That's the point. The device being compromised to the point that malicious code is actually meddling with the bank app is the only time that having it fail attestation would be useful. The other cases are useless/vexing false positives. But attestation doesn't happen in the one case it would be useful because then the attacker-controlled code won't even attempt to do it, it will just exfiltrate the user's credentials to the attacker.

You aren't responding to the scenario that was posed. You're assuming an isolated compromised app on an otherwise clean device. GP is assuming a compromised device.

Of course attestation does nothing to improve the "single compromised app" case since (assuming Android) that goes nowhere either way. The only thing attestation does is meddle in end user affairs.

>if the phone isn't compromised then the owner having root isn't a security problem

The scenario is the phone isn't compromised. Having root means you, or an app you run can bypass the security protecting the authentication token.

And now you're being intentionally difficult. Please interpret things in the most plausible manner. Beyond common decency, it's part of the site guidelines.

By "not compromised" GP clearly meant a scenario where no malicious apps are present.

I agree that's a serious omission. I responded to your scenario (a nonzero number of malicious apps) in my earlier comment. Any Android device will defend against that regardless of the presence of attestation.

Any non-android device can still use online banking and thus attestation doesn't appear to accomplish anything legitimate. Building out proper support for hardware tokens would provide superior security in approximately all cases.

The specific "root on android" scenario isn't generally a concern. Typical implementations require explicitly granting the capability to a given app. A malicious app can't gain it without fooling the user, at which point it could more easily phish the credentials and possibly even proxy an entire session.

>Please interpret things in the most plausible manner.

Your suggestion is not plausible as every security feature has 0 security value if there is nothing malicous. It would be like someone saying that antivirus is useless because if someone doesn't have a virus then it doesn't do anything.

>Any Android device will defend against that regardless of the presence of attestation.

Rooted android devices can be set up in a way that malicous apps can gain root and then read it.

>Any non-android device can still use online banking

But this comes with a different risk profile. A bank can reduce risk for a subset of their customers.

>Building out proper support for hardware tokens would provide superior security in approximately all cases.

I think usually the hardware token gains you access to an authentication token. You don't sign every request you are making with a hardware only key.

>Typical implementations require explicitly granting the capability to a given app.

And the majority of users have no clue what an app is able to do. If root is given to it then it can do anything. This is in contrast to when root isn't available and users are protected by the sandbox the app is in.

Yes, a user could intentionally do something stupid. A user could also voluntarily wire the majority of his balance to a foreign account. That fact has no bearing on a discussion about the security impacts of attestation.

From a technical perspective device attestation doesn't add anything here. The typical user doesn't receive any additional protection. All of its supposed "benefits" ultimately stem from the restriction of end user choice but those same end users _already_ have a practically limitless selection of stupid choices available to them. And these are generally very deliberate choices that we're talking about here. Not subtle confusion.

> the majority of users have no clue what an app is able to do. If root is given to it then it can do anything.

If a user is stupid enough to seek out root, ignore all warnings, install a malicious app, and explicitly grant it root, then he was fated to fall for a much simpler scam regardless. Such as granting a malicious app admin on his laptop and then logging into online banking. Or installing a phishing app that proxies the session to the bank.

Typical users don't seek out root, don't install custom ROMs, and don't consent to the warning message about installing APKs from unknown sources. Grandma isn't ricing her mobile phone.

Support for hardware tokens or TOTP would address widespread real world attacks. Shutting down various customer service social engineering account reset tactics would also help, but that would actually cost money. In contrast attestation doesn't accomplish anything other than infringing end user choice. It's a highly unethical waste of resources.

> It would be like someone saying that antivirus is useless because if someone doesn't have a virus then it doesn't do anything.

Suppose you have an "antivirus" program that works like this: The system makes a list of every program that runs as root during boot and then at the end of boot the antivirus program checks the list for unexpected programs and if it finds any it displays an error and refuses to display the login prompt to prevent the user from typing their password into a compromised device.

That "antivirus" system is useless, because if a malicious program did run as root during boot then it could just reconfigure the system to display the login prompt unconditionally.

The way attestation is nominally supposed to work is that a remote system cryptographically verifies the state of the local machine before giving it some secret. But that doesn't work in this case because the secret -- the thing that allows the user to sign in -- is coming from the local user rather than a remote machine. The attacker doesn't need to perform attestation or retrieve anything from a remote machine in order to display a local login prompt and collect the user's credentials, and that's the end of the game.

> Rooted android devices can be set up in a way that malicous apps can gain root and then read it.

So can PCs, or various non-rooted android devices that bank apps run on even though they have known unpatched vulnerabilities.

> But this comes with a different risk profile. A bank can reduce risk for a subset of their customers.

How is it reducing risk for anyone? Each person still has the same risk profile as before. The person with a locked Android device still has a locked Android device, the person with an unlocked Android device is now forced to use a PC which is an inconvenience with no security advantage.

Just because an antivirus isn't perfect that doesn't mean it's useless.

A bank may not want to put in the extra effort of supporting 3 platforms of secure phones, insecure phones, and insecure web and would prefer to support less and potentially dropping features from the web platform.

> Just because an antivirus isn't perfect that doesn't mean it's useless.

If the security of your system depends on the attacker not doing something the attacker can easily do, it's useless.

> A bank may not want to put in the extra effort of supporting 3 platforms of secure phones, insecure phones, and insecure web and would prefer to support less and potentially dropping features from the web platform.

So now we've gone from some kind of hypothetical security benefit to "banks don't support that because they're lazy". Except that adding support for attestation is work they do for no reason, because the people using locked phones are still using locked phones and preventing people with unlocked phones from using them instead of the web page is paying to do work only in order to cause customers trouble.