Grammarly shared its tokens with all websites
bugs.chromium.org
bugs.chromium.org
> "By uploading or entering any User Content, you give Grammarly (and those it works with) a nonexclusive, worldwide, royalty-free and fully-paid, transferable and sublicensable, perpetual, and irrevocable license to copy, store and use your User Content in connection with the provision of the Software and the Services and to improve the algorithms underlying the Software and the Services."
First I thought that may be this is just me being paranoid. So I compared their terms of service with Evernote's[2] and summarize the differences for them in the support ticket asking termination of my account. I reproduce that here for reference:
> In Evernote's TOS
> – It's clearly noted that the user retains their Copyright to the content;
> – and the license to Evernote is a limited and they don't "obtain any right, title or > interest" other than they point out;
> – they don't require sublicensable and transferable rights to User Content which is different than having the rights to share the content with other contractual partners (which they require);
> – the agreement on content is irrevocable as long as the content is stored on the service.
[1]: https://www.grammarly.com/terms
[2]: https://evernote.com/legal/tos.php
Edit: Formatting.
Collecting everything you type into a web browser (or MS Office) and sending it to them seems like a really bad idea.
Of course, in the case of Microsoft or Google, you presumably either have disabled the setting or you place your trust in their security practices that it is okay, because they are top tier companies, and most people send them all their private data anyways.
There are a LOT of things out there that collect everything you type these days, and rarely to people want to define them as keyloggers.
When people want privacy they will inevitably have to give up usability. I ditched Swiftkey for an open source Android keyboard that doesn't connect online or asks for any permissions. Its CRAP but it doesn't leak.
It has to work this way or browsers wouldn’t be truly extensible. Be mindful of which extensions you install.
And you only can spare both the time and cognitive load to do this for at most a dozen or two applications, if you really care. The rest, you just have to trust that enough other people are watching carefully.
But the average person isn't going to keep up with even one application. They only bought their computer so they could browse the web and check emails, not so they could learn the details of how it works.
Likewise, most of us don't buy a car in order to spend a lot of time learning about exactly how a combustion engine works. We don't have the time.
Granted, this board is laden with engineers who will make the time to understand how their tools work, but we simply cannot expect this kind of effort from most people.
So, like we have to trust lower-level components to be scrutinized elsewhere, and trust we will be alerted in case of critical issues, the general population must trust the "nerds" to get things right and keep them safe.
This means that typically, the best attack surfaces will be small, widely-distributed, low-level software stacks whose developers can easily be compromised. Not just software either, but hardware.
It does seem like this is ultimately a battle we are going to lose without regulatory legislation in domains that require mass-deployment of software that can potentially breach Constitutional rights. In order to be federally qualified as "privacy-friendly", you have to meet certain guidelines both on a hardware and software level. This would include automatic transmission or collection of certain kinds of data without very express permission.
That sounds an awful lot like "Be careful what email attachments you open." Blaming the user never worked out then either.
I guess android's "partial permission" is the right thing...
https://languagetool.org/ for one supports running your own instance of the server-side portion out-of-the-box. You could run it truly locally assuming your device is appropriate, or your own server which might be more flexible as you can support a greater range of devices and share custom dictionaries between them.
Gives me a "privacy policy", "continue and don't ask again" or "cancel" hyperlink, nothing else. If I want to use it the first time.
But if you open the extension options under the "more tools" > extensions tab you can set the LanguageTool API server URL[0].
Going to try to setup a server over the weekend, thanks.
Yes... But your browser won't detect "spellcheck do can browser your"
Grammar is more than just spelling.
Trying I promise, even different languages. But I'm not a native speaker and a bit dyslexic.
So someone or thing looking over my shoulder would be nice.
It's eye-opening to people when I ask them about an extension they have, say "Honey", and they say they like it because it saves them money. And then I point out it can access everything they do online, and ask them if that's a concern or not.
Chrome team, if they were security-focused, would not permit any closed source extensions which have access to all website data.
People don't seem to understand sometimes that if an extension has this sort of access, you need to be able to trust your browser extensions as much as you trust your browser itself.
I am curious how you manage your passwords.
The notion that it's a good idea to trust a browser extension for secrets management is pretty bizarre to me if you're protecting high value assets.
(Not directed at you personally, but I often hear such comments from people who are then perfectly fine to use a password manager in X11, where in a the default configuration every application can read your keystrokes, screen grabs, clipboard, etc.)
[1] Preferably one that communicates with an out-of-process password manager over an authenticated channel like 1Password.
Beyond that, the "never reuse passwords" adage is horribly oversold. If it handles my money, my email, or my web hosting, it needs to be unique. Passwords for places I comment are commonly reused and not as sophisticated because it is not seriously impactful to me if someone gets a hold of them.
Reuse passwords for sites that can't meaningfully harm you if they get compromised. Minimize how many accounts can harm you by not saving your credit card info in most of them, uncheck that box when you pay for stuff.
I'm also insanely liberal about deploying 2FA. I have it everywhere it's available, even sites with common/stupid passwords. So a lot of sites I don't bother with unique passwords will still be somewhat protected if my password is compromised. I'm also subscribed to haveibeenpwned with every email address I've ever used for anything.
"LastPass encrypts your Vault before it goes to the server using 256-bit AES encryption. Since the Vault is already encrypted before it leaves your computer and reaches the LastPass server, not even LastPass employees can see your sensitive data"
If there is an attack still possible (even using LastPass employees) can you post it here?
https://bugs.chromium.org/p/project-zero/issues/detail?id=88... https://bugs.chromium.org/p/project-zero/issues/detail?id=11... https://bugs.chromium.org/p/project-zero/issues/detail?id=12... https://bugs.chromium.org/p/project-zero/issues/detail?id=12... https://bugs.chromium.org/p/project-zero/issues/detail?id=12...
At a glance what they have in common is flaws in the scripts that the LastPass extension injects into pages. The injected scripts can communicate with the extension core with a set of RPCs. Each of these issues is a way of tricking the extension into running RPCs from untrusted JavaScript on any web page. The RPCs available allow an attacker to fetch the credentials for any site in the database or even execute arbitrary code on the host.
The fix for that was to not use autofill and revert to manually grabbing your username/password when filling out a login form.
Aside from that, I am not aware of other "hot" attack vectors.
Users still need to practice skepticism and ultimately it is their responsibility to protect their passwords. But LastPass has been a very good citizen when it comes to being as secure as possible.
Same as Google/Firefox autocomplete and history, or keyboard spell checker, or email autocomplete and spell checker, etc.
So linux and android are also in the boat of having apps that make your life easier, also need security enforced.
Just wanted to say, its not a Windows only issue, OSX, iphone, android, they are all going to be affected to simular issues.
> Grammarly had fixed the issue and released an update to the Chrome Web Store within a few hours, a really impressive response time.
Nice to see a company take this kind of thing appropriately seriously (although of course it should never have happened in the first place).
This prevents users from creating click bait headlines to get that karma tho, the majority of the time. Cases where the actual title is the click bait like this one, are the unintended consequence of that policy.
My annual sub end March 1st which is about perfect timing since they were trying to double the annual rate without proper notice, in my opinion.
Turns out Grammarly was injecting HTML into the editor, which in turn was being included in the email body!
For better or for worse, between browser extensions loaded by the end user, and "tags" injected by your well-meaning business analytics team - see my comment here: https://news.ycombinator.com/item?id=16314501 ), the extension ecosystem has become the new Internet Explorer in terms of compatibility testing. Luckily most of the workarounds are trivial, but it's essential to have good QA on actual client machines if you're doing, well, anything at all.
Or eventually web developers will start putting that stuff in by default, bootstrap will come with it, etc.. and these companies will see less and less traffic, and they'll start coming up with reasons to ignore it.
I dug in a bit, and it turns out that Grammarly was embedding a gigantic amount of code into the email messages in the form of stylesheets and other things.
Needless to say, after raising it up the chain, we had the extension blocked company-wide.
That said, I removed it because the button in the lower right kept getting in the way of things like resizing the textarea.
To get to this point, I did primarily two things: I read a lot, and I wrote a lot. I then submitted my writing to my peers, similarly interested in creative writing, who would also - in addition to the story, characters and so on - criticise my word choices and my grammar. It took years, but it was enjoyable, and I acquired my second (and last, after programming) skill that could be of some value.
A couple of years later, when I started working, I had to switch to English. It wouldn't be that big of a deal were it not for my experiences: every time I had to write anything, I felt incredibly constrained, like I'm missing half of my brain. I was used to being able to express myself precisely, clearly and elegantly in writing - all of that stopped working after the switch. It's incredibly frustrating, to the point that for a few years I was in complete denial and refused to write in English wherever I could get away with it.
Well, I thought, I got proficient in Polish, so technically I should be able to get to the same point in English, right? It's easy - I just need to read a lot and have a group of people who'd like to read my writings and correct my mistakes. Easy!
...however, I'm not in high school anymore. Between work and the little social life I have, there's not that much time available for pursuing other matters. I do read a lot, exclusively in English, but these are mostly tech-related articles, blogs, and books, written by people who couldn't care less about beauty and elegance of their writing. It's actually counterproductive if my goal is to get better at writing - such posts are chock-full of both errors and merely weird wordings and constructs. And nobody seems to care.
There is a creative writing StackExchange (and many other places), where I could submit my texts to get the criticism and corrections I need. Unfortunately, I don't have the time - even if I had the skill - to do my part of the deal, that is, to read and comment on writings of others. I'd feel bad exploiting strangers like that.
As you probably already guessed, this is where Grammarly comes up. It gives me a bit of the feedback I need to improve my writing. It's not at the level of other humans, which is obvious, but it does catch some mistakes and some stylistic problems. It doesn't rely on unpaid work of others, so I have no qualms about using it. I'm not worried about following its advice because even if it's wrong, nobody would care. The amount of contempt for the language in the tech community is staggering; average tech-related writing is on such a level that I'd rather chop my hand off than write like that, but it shows just how unimportant correctness and elegance is for people (as long as it gets the point across... right?)
So, to summarize and get back to the topic at hand: Grammarly is non-ideal on so many fronts, that to simply enumerate them would take until Friday (it's Wed today). But it's also the only tool I can rely on, and it does an acceptable job at what it does. It breaks web pages, it's unusable from outside a browser, it's error indicator is frequently displayed 3+ lines from where it should be, it's stupidly dumb and cannot, by itself, tell where the additional "actually" is actually needed, but it's the only help I can get, so I use it.
Of course, if your goal is merely communication you do not need anything other than basic spell-checker and a book on basic grammar. On the other hand, if your goals are similar to mine, then Grammarly is one step above that combo. It isn't, and probably won't ever be, anywhere near the level of human reviewers, but it is something.
PS. For a long time, I wanted a feature in Grammarly that would automatically slurp content of blog posts and articles in, so that I can just click a couple of times and then read the post without all the mistakes and weirdness people so often put in there. Reading the top-voted comment here, about the license, I see why they won't implement it. Similarly, I guess Emacs plugin is not going to emerge anytime soon. Whatever you use Grammarly for, you should assume it's public. It doesn't make it any less usefull for writing comments on some sites or posts for my blog, though.
Relying on "best practices" is always a security disaster waiting to happen, if you don’t enforce security and separation in the design of the APIs and languages already, you won’t get security.
I don't get this thinking at all. Browser addons are trusted. That's the point - they have special privileges to adjust browser behaviour.
If you go around installing malicious addons, you get no more sympathy from me than if you'd gone around installing malicious kernel modules.
For example, Firefox’ screenshot addon would inject HTML into the page, and then the page could take the screenshot’s data and use it.
Addons currently have no way to reliably display their own UI on top of the page, without the page intercepting it.
I agree, addons' workings shouldn't be exposed to untrusted websites.
The extension permissions API already offers enough restrictions. As a user, I simply do not install extensions that need access to all pages, or I only enable them on pages where I need them.
It’d be much more secure to allow the extension to do that without requiring it to put this into the page.
For example, currently you have the DOM of the page, and the extension’s UI is injected into it.
It would be preferable if you’d treat each tab as if it was like
<tab>
<browser src="actualpage"/>
<overlay src="extension1"/>
</tab>
I think you could use the Shadow DOM in closed mode to prevent any information from leaking. [1]
[1]: https://blog.revillweb.com/open-vs-closed-shadow-dom-9f3d742...
Ideally you’d want to allow addons to modify pages in a way that pages can not detect or interfere with.
(e.g., some newspapers used to run JS to remove the AdBlock "block this" UI whenever you tried to remove an ad)
An extension can stuff its UI within an iframe that the host page can't manipulate, but that does come with some UI limitations.
If it’s a message in the browser chrome itself, and not as <div> in the page, it is possible to add distinguishing UI.
With the current system, it’s literally impossible to distinguish the UI
When you add an extension page script, you get access to a page's DOM, but you're completely isolated from the page's own JS: You get your own JS context and window object without any modifications the page may have done to it. That's usually reasonable as a page can mess with the built-in methods of its context, so if an extension were to rely on them, the risk of privilege escalation attacks would be really high.
Except sometimes an extension does want to interact with the page JS, (e.g. for accessing data the page only keeps in JS objects but not in the DOM.)
As far as I know, there is no safe way to do this in Chrome. The recommended (!) way is to inject a script element into the DOM and exchange data with the page script via some makeshift communication channel, e.g. postMessaging yourself. This will of course drop you right back into the hall-of-mirrors of potentially manipulated builtins that the page script isolation was trying to keep you out of. But apparently now it's ok if you have to deal with that by yourself.
From the looks of it, it seems Grammarly tried to open exactly that kind of communication channel and didn't correctly secure it.
Chrome and Firefox have _very_ different behaviors here.
In Chrome, there is no way at all to get hold of the page's JS objects.
In Firefox, the default behavior is that you don't interact with them, but you can explicitly ask for them and once you have them you can work with them. Depending on what you do with them, you may or may not be creating security bugs, of course.
Critically, in Firefox you can have your separate clean builtins _and_ be interacting with actual page JS objects at the same time.
There are arguments for and against both models, of course.
you can download it and run it as stand alone application
I'll close the ticket once the item have been completed.It's a choice between visible pain and invisible weakness. The early disclosure hurts and amplifies the issue. However, the malicious actors you need to worry about would be glad for every day without disclosure, because every day without disclosure is profit to them.
thx!
Enable developer mode, and then Update Extensions Now. That forces a manual update (although the version you're referring to appears to be the latest).
1) Grammarly is a fancy grammar/spelling correction tool.
2) You use it by opening an account, and installing their browder extension.
3) As you type text into a web page, the extension sends that input back to Grammarly, where their software analyzes it and provides correction recommendations.
4) The text that was sent back is persisted under your account, and is available for retrieval.
5) A software bug in their extension allowed a script on any site to see your Grammarly auth token.
6) As the result, any malicious site could log into your account, and see what you've been typing.
Is that the rough gist of it? If yes, then how in the world does #4 make any sense? Why store and expose that data, knowing that it's likely to contain troves and troves of sensitive and PII information...
It's only mentioned in the comment, but the amount of mistakes in tech-related writing, by native English speakers and otherwise, is gargantuan and overwhelming, and I'd wish using something like Grammarly (if safe, open source and so on) was a requirement for putting your writing on the web.
As it is, the quality of writing is so bad, that I (as a foreign speaker) don't improve my English in any way by reading it, and I have to be very careful not to repeat these mistakes in my own writing later.
It looks like a good product. If they offered a true offline version for desktop with 4+ updates a year, I could see myself paying for it.
How long are those tokens valid for?
Did they invalidate all existing tokens?
Using Grammarly is stupid, paying them is downright insane.
https://support.google.com/chrome/a/answer/187202?hl=en
This is one place where Google actually did really do their homework, IMHO.
Most software patching solutions have a way to push files though, and you can use something like this: https://www.itsupportguides.com/knowledge-base/tech-tips-tri...
I recently loaded Wordperfect for Win3.1 into a Win3.1+Dosbox instance because I remember it's grammar checker back in the 90s was far superior to what's built into MS Word today. I've been meaning to test it out and do a comparison blog post.