That is precisely what these devices are designed to stop. The device has a private key stored in hardware in a way that it cannot be retrieved by software. When you use one of these devices you dramatically decrease your number of attack vectors because now the attack has to happen physically. Someone has to actually steal your physical key. And because this is your "second factor", if that happens they still also have to have your password.
Two-factor apps get closer to this, but they usually can be copied. For instance, 1Password can be set up to mimic Google Authenticator.
> I do want to know if this is purely marketing hype
This is most definitely NOT marketing hype. It is the current security best practice.
I was storing backup codes in 1P before I realized that I was putting all my eggs in one proverbial basket.
Probably wouldn't be a bad idea to have a distinct password vault for TOTP seeds, although they'll be stored on whatever device you generate codes from anyway (hopefully in a secure way). At the very least, it might be helpful for moving between TOTP devices without needing to do the random steps required by different services!
It's pretty obvious why this changed, however, it is a major increase in exposure and a terrible change overall. It is discussed in [1] and mostly the answers aren't very satisfying as demonstrated in [2].
[1] https://discussions.agilebits.com/discussion/101560/secure-m...
[2] https://discussions.agilebits.com/discussion/101551/article-...
I guess it makes sense. By the time an adversary gets your key you would of noticed and have locked that key from your account.
With SecurID and similar technologies vendors technically didn't need to retain the secrets inside those devices after they'd been manufactured and shipped, but you can see the practical temptation.
On a smaller scale, since the system doesn't use a shared secret you should just swap a brand new Security Key with somebody else or if deploying to an organisation just muddle them and let people pick whichever one they want. You don't care which key you have, the more random the better.
Definitely, it's an expensive vulnerability to simply reveal though. YubiKeys are proprietary so they can't be audited by non-contracted third parties. There are FOSS security keys that don't have that problem though.
1. They can steal the secret, from you or from the other party (the service you're authenticating to)
2. They can relay your answer, without knowing the secret and since the answer is genuine their relayed answer will be accepted. Example you type the One Time Code into https://fakebank.example/ thinking it is your bank, then the bad guys simply paste that same code into https://realbank.example the actual site of your bank and it works - they get into your account.
FIDO Security Keys don't use a shared secret, instead they use public key crypto. Your Security Key proves it is still the same key as it was when you registered, but the remote site doesn't have any secrets so they could even _publish_ the data they store to authenticate you and that wouldn't matter to the system's security.
And it relies on the web browser to tie the FQDN into the data signed. If you press the button on the Security Key on https://fakebank.example/ the results are only proof for fakebank, operated by the bad guys, if they forward it to https://realbank.example/ it won't work because the name is wrong and it doesn't match.
Does that help? There are a bunch of other cool things about Security Keys, but those are the main ways it's simply better for security than a conventional TOTP-style second factor.
For example, Google reported that deploying U2F blocked all phising attempts of their employees : https://krebsonsecurity.com/2018/07/google-security-keys-neu....
evil.com can still claim to be mysite.com and trick users into using their security key there, no?
It would presumably have to trick the user into "registering again" since there is no valid key handle for FIDO2 (not sure about classic U2F), but I doubt that users would notice the difference in flows.
The real win is that this spoof would still not give evil.com a credential scoped to mysite.com, but a user would be none the wiser that they are on a different site than they expect (which can still be problematic depending on the nature of the site).
Even if evil.com gets you to register and present your key, they cannot forward it to mysite.com. Even if they go to all the trouble to completely mitm you and the site looks identical to mysite.com, they cannot get the emails or get it to send money.
This is because evil.com cannot pretend to be you when they interact with mysite.com no matter what you've given them.
It's going to be hard to trick me into thinking I've logged into my gmail if none of my emails are there!
Unless they somehow convince you to put your yubikey in the mail and physically send it to them...
I'd just be careful about overly relying on this property or calling it anything like mutual authentication:
If an attacker can make an educated guess about a user's account contents, they could still convince them to provide additional personal information once they let their guard down after authenticating.
I have never seen a solution like this, but I can't see how it would avoid simple relay phish, it seems as though bad guys can just get themselves a legitimate encrypted barcode and have their victim scan it.
Maybe you haven't explained this solution properly, do you have a link?
Since the Yubikey (and friends) operates by generating a private key on the device itself, inaccessible to software, it means that the key is hard to clone. They'd have to steal it off your person. Combined with generating a unique key pair for every website and checking that the website is what it claims to be, this means that the key is resistant to replay attacks, which number based authenticator apps are vulnerable to.
That isn't to say that it's immune to phishing, but it's much, much harder to phish someone who is using a Yubikey because an attacker would need to compromise someone's DNS configuration or SSL configuration to impersonate the website, at which point... why are they bothering with phishing?
Yubico has some FIPS certified devices[2], which means that they've presented a design that shows the device has mechanisms to prevent secrets from being extracted, and they're only using algorithms known by NIST not to leak secrets.
> Also pardon me if I confused two-factor as the Google Authenticator app.
Multi-factor authentication is about managing risk, and discussions about risk are naturally fuzzy and vague.
I'll try a concrete analogy; consider firearms safety.
Some typical rules[3]: 1. keep the weapon pointed down range at all times, 2. keep your finger out of the trigger well, 3. treat the weapon as loaded at all times.
Each rule is a factor, and to accidentally hurt someone you have to violate all the rules at once.
Multiple factors work best if they are orthogonal, that is, when a given action results in only breaching a single factor. That's why factors tend to be phrased as "something you know," "something you are," "something you have".
The authenticator app and a Yubikey are doing the exact same thing: they're establishing the "something you have" factor.
Since the two factors work when an attacker must both obtain the device and get your password, if your phone has both passwords and authenticator apps, the additional factors aren't minimizing that risk.
[1]: The automatic vendor lock-in makes it a great business model...
[2]: https://www.yubico.com/business/product/yubikey-fips
[3]: There are many more, but take a class on it rather than depend on the Internet.
At least two HW backdoors have been found in chips such as this one in earlier versions, the first using an improper RSALib RSA variant (developed by NSA or Mossad). The source for the second, SCADA, is unknown.
So any nation state wanting the most people to use it, will advertise it as such. "unbeatable security and can protect against a variety of threats, including nation-state attackers."
This happened with the Infineon backdoor, the SCADA backdoor, the earlier Crypto AG backdoors. https://acmccs.github.io/papers/p1631-nemecA.pdf
Obviously they are biased – they are in the business of selling hardware, not software, but their reasoning still makes sense to me.
It's arguable whether "destructive updates" that invalidate all existing registrations should be allowed, but that could have its own problems in terms of availability if not clearly communicated.
On a related note, I wonder if there is actually a more secure way of preventing illegitimate firmware updates in JavaCard/GlobalPlatform than randomizing the card manager keys, which is what Yubico does for the NEO: https://www.yubico.com/2016/05/secure-hardware-vs-open-sourc...
This should be impossible.
So really there's no point in cloning. Straight up theft is the bigger concern.
Alright, I'm not a security expert, but I'm not completely illiterate to basic computer security. Anyone care to chime in how this is much more secure than a two-factor app?
Sure there's the obvious, nobody can just copy the two-factor app off my phone with all the codes and have the same codes (I've upgraded phones and taken my codes with me before...) but whose to stop someone from cloning a Yubico key?
Again I'm not an expert, but I do want to know if this is purely marketing hype or if there's some security to Yubikey and friends that I'm not aware of.
Also pardon me if I confused two-factor as the Google Authenticator app. Too many "factor" terms get a bit confusing after a while.