In-App Browsers Considered Harmful

How many apps on your iPhone or iPad have a built-in browser?

Would it surprise you to know that every one of those apps could eavesdrop on your typing? Even when it’s in a secure login screen with a password field?

Here is a proof-of-concept (ZIP file) that shows how an app can do this. For those of you who don’t have Xcode installed, here’s a video that shows what’s going on:

A few things to note about what you’re seeing:

  • The information at the top of the screen is generated by the app, not the web page. This information could easily be uploaded to remote server.
  • This is not phishing: the site shown is the actual Twitter website. This technique can be applied to any site that has a input form. All the attacker needs to know can easily be obtained by viewing the public facing HTML on the site.
  • The app is stealing your username and password by watching what you type on the site. There’s nothing the site owner can do about this, since the web view has control over JavaScript that runs in the browser.
  • The site content is also modified: the text on the button label is normally “Sign in” and has been changed to “SUCK IT UP”. It seemed appropriate.
  • This technique works in iOS 7 and 8 (and probably earlier versions, but I didn’t have an easy way to test them.)

OMFG APPLE IS HACKING ME

No, this is not a WebKit bug.

The Shadow DOM does a great job of protecting static user content on a web page. It’s not possible to use JavaScript to view the contents of an input field on iOS since the current value attribute is actually being held in a platform-native control. The value of that control is uploaded when the user submits a <form>.

I don’t know for sure, but I suspect that the keyCode attribute of the KeyboardEvent in the JavaScript event handler is provided for backward compatibility. This API has been deprecated but there are still plenty of web pages out there that use it to handle keyboard input.

In fact, both the techniques shown in the sample app can be used for good as well as evil. Changing the content of a web page is a good thing when it’s done to make a page more readable or accessible. Handling keyboard events can also guide a user through a complex form or make viewing a slide show easier.

These are not inherently bad web technologies. The problem is that an iOS app has as much access to these technologies as the developer of the web page.

OAuth To The Rescue. Or Not.

Websites have been dealing with username and password attacks for as long as there have been <input> fields on their pages. One of the primary goals of OAuth was to keep a user’s login information away from an external website or app.

OAuth does this by exchanging cryptographically signed tokens between the site where the user has an account and the app or web service that wants to access that account. A key factor in making this secure is that the exchange of these secure tokens is done through a trusted channel: the user’s web browser. Twitter has required third-party developers to use OAuth since 2010.

As early as 2008, the developers of OAuth recommended the following:

We’re trying to ensure that users are only exposed to the safest way to disclose their location using OAuth. To do this, it’s critical that a fundamental principal of browser-based authentication is followed; that the contexts of the third party application and the web service authentication remain separate. To allow users to grant trust to an application, they must perform the OAuth action within their web browser, not within the applications themselves. Otherwise, there is no way to verify the identity and authenticity of any page which asks for their username and password. Users must not ever enter their username and password into a third party application when a browser-based authentication API like OAuth is available.

There is always a tradeoff between usability and security. Doing the OAuth token exchange with an in-app browser makes it easier for a user to login, but they’ll have no idea if their personal information was captured. That is why Twitterrific did its token exchange in Safari, even though it’s a more complex user interaction and a more difficult technical implementation. As a user, I know that there’s no way for my login to be compromised when the transaction involves Safari.

Unfortunately, Apple’s current App Review policy does not agree with this recommendation or with Twittterrific’s previous implementation. This is why our update for iOS 8 was delayed—it was the first time since the launch of the App Store that we haven’t had a new version on release day.

(Apple folks can learn more about this situation by reviewing Radar #18419943)

Recommendations for Apple

Apple has taken a strong and welcome stance on privacy. They’ve recently been implicated in some high profile attacks so they definitely have skin in this game. Hell, they even want to protect us from the US government watching what we do online!

There’s no denying that the behavior demonstrated above could be very harmful in the wrong hands. It’s also Apple’s job as the gatekeeper for iOS to keep malicious apps out of the App Store. But how?

I don’t think it’s feasible to catch misbehaving apps at review time. There are a huge number of apps that need to be reviewed every day, especially when new versions of iOS are released. Many of these apps use in-app browsers which would require extra time and effort to vet. Longer review times benefit no one: developers, Apple and our customers need timely updates.

It’s also very easy to an app to hide any nefarious activity. JavaScript has an eval() function that makes it easy for code to be obfuscated and very difficult to be checked at review time. Look at this page and see if you can guess how the uppercase text was created. Then view the HTML source and see how wrong you were.

Additionally, an app that wants to collect your information can easily implement a remote switch that disables the functionality while the app is in review. App reviewers won’t stand a chance.

Changing how WebKit and UIWebView behave isn’t practical either. To prevent this keylogging technique, Apple would need to release a new version of iOS for each version that included Safari and WebKit. Do you really think they’re going to do a point release of iOS 3?

And this brings me back to protecting users with OAuth. It’s designed to avoid these problems and works well to maintain privacy. Granted, it goes against section 10.6 of the App Store Review Guidelines, but in my opinion, this is a case where user security trumps usability. Apple should change their policy for apps that use OAuth.

Recommendations for Users

Another goal of this essay is to increase user awareness of the potential dangers of using an in-app browser. You should never enter any private information while you’re using an app that’s not Safari.

An in-app browser is a great tool for quickly viewing web content, especially for things like links in Twitterrific’s timeline. But if you should always open a link in Safari if you have any concern that your information might be collected. Safari is the only app on iOS that comes with Apple’s guarantee of security.

(For the record, we never collect any private information in any of the Iconfactory apps. And we never will.)