Two side notes: 1) in the modern multi-GPU era, no knowledgeable attacker will use rainbow tables for fast hashes like MD5 or SHA1. 2) 100+ length is massive overkill, even for a fast hash. The capacity of the entire Bitcoin network can't crack a randomly generated, upper/lower/number/special 20-character password in 100 years - even assuming the continuing validity of Moore's Law! twitter.com/jmgosney/status/714599158229786625 . Always do the math. 95^100 is ... much bigger than you might think.

1) Oh yeah that's quite true actually, they have enough raw power to calculate billions of hashes every second so I guess rainbow tables is moot for them

2) I agree yeah it is overkill, but it was more in the frame of choosing a password which would pretty much never be cracked in one's lifetime to reuse.

Well ... 20 would be "multiple human lifetimes" overkill. 100 isn't just overkill - it's "the time it would take to reach the entropic heat-death of the known Universe, billions of times over" overkill. It's the mathematical equivalent of rolling a pair of six-sided dice and hoping to get 17 trillion. Follow the math in the Twitter thread and you'll see what I mean. It may seem nit-picky, but other folks reading the thread need to develop an intuitive understanding of just how out of scale a 100-character password would be.

(And honestly, even 20 is in the powers-of-millennia range: 95^20 is ~3.584 × 10^39 wolframalpha.com/input/?i=95%5E20)

No matter how complicated your password, it might be cracked on a brute forcers first try

@RoyceWilliams I agree, it is out of human range, but if it is feasible for someone to use there's no harm right? We don't know whatg kind of technology can me realised in our time, nice to be completely safe, like over-packing, within reason.

@JCRM The probability of that though is probably worse than throwing 100,000 coins in the air and having all of them land on heads, if it is a 100 character password which is made randomly and has a crazy amount of entropy.

@Azxdreuwa Beyond pure math, there are upper bounds also derived from our current understanding of physics - specifically: en.wikipedia.org/wiki/Landauer's_principle ... so if cracking a 100-character password becomes possible in our lifetimes, it would have to transcend most of what we know about the fundamental nature of the universe.

Nobody is going to waste years trying to bruteforce a password. If it can't be guessed in a relatively short time, then they will use other methods to steal it. Also, for authentication and authorization on a website, I don't think it even needs to take account of future changes in technology: if it ever becomes weak in the future you can always change it. Also, I don't think you will easily remember a 100-char random password, you'll have to write it down (yet another way to steal it) or use a password manager (and that will help you always use different passwords).

@Azxdreuwa, indeed it is, but the odds of that are exactly the same as ever other combination (not permutation) - and one of those will happen every time 100,000 coins are thrown in the air. A strong passord makes brute forcing improbable, but doesn't prevent it being cracked quickly.

@JCRM Yes, it does - on average. That's the way randomness and statistics work. If what you are saying was true, there'd be no point in increasing the length or randomness.

@RoyceWilliams you increase the length and randomness to make it less likely - but you don't treat it as uncrackable and reuse it despite it having been on a compromised site. AKA "someone wins the lottery every week."

Fun fact: authentication is most often done plaintext these days. That is, you tell server "psst, my password is correct horse battery staple". If server is compromised, getting your plaintext password is trivial.

@Azxdreuwa "The probability of that though is probably worse than throwing 100,000 coins in the air and having all of them land on heads ..." - You're a little optimistic there. A 100 character random password from [A-Za-z0-9] has just under 596 bits of entropy, so the probability of guessing that password on the first try is better than throwing 596 coins in the air and having them all land on heads. Still really unlikely though.

@marcelm I'm not in your sense, the password I had in mind has an a much much greater entropy, and consists of many more symbols than the character and number alphabet you worked with.

Wow I've seen at least 3 different XKCD's referenced on this page and I didn't even read to the bottom. That might be a new record, even for infosec.SE.

@el.pescado And that's not even taking into account the fact that some organizations store the passwords as plaintext. The fact that only an idiot would do it that way doesn't help much, since the stupider the sysadmin, the more likely the system will be among those compromised in the first place.

@Azxdreuwa That won't help you much. If you include all readable ASCII characters, you'll only go from log2(95^100) ≈ 596 bits (coins) to log2(62^100) ≈ 657. Barely an increase.

@marcelm Over three quintillion times as many possibilities is barely an increase?

So, all that would be necessary to steal your great password is one site that stores it in plaintext or accidental usage in a malicious login form that steals passwords? Remind me why you thought this was a good idea.

@iheanyi I never called it good though lol, I just wanted to understand the risks behind using such a scheme to log into websites through secure tunnels, but it's led me to understand much more about the risks of password reuse in any situation.

@Ray 61 bits gained, compared to the 100,000 bits Azxdreuwa claimed initially? Yes, that's barely.

@marcelm The list of unicode characters is greater than 136,000. Added with all symbols from the sets [a-z],[A,Z],[0,9],[ASCII - ([a-z][A-Z][0-9])], I truly feel the entropy is much much greater than what you are making it out to be. There are simply too many combinations of symbols, which is why I likened it to throwing 100,000 coins and expecting heads from each one, this is simply verging in the realm of impossibility unless by some infinitesimally small chance that you somehow manage to.

The caveat being that whichever website one uses supports the set of unicode characters of course...

And you can rule out the possibility that an attacker can get the cleartext password from any of these sites?

@PlasmaHH I was more so originally thinking that this password would only be used on websites that have a clear layout of their security practices, not really just those run of a mill websites which will store it in cleartext, since, at least in the case of large companies such as Google and Microsoft, there is simply no way they would store passwords in plaintext, Microsoft especially, they work directly with NIST!

@Azxdreuwa: There are more ways to get to the passwords than just storing them in plaintext. The servers themselves could get compromised.

@PlasmaHH I do see what you mean, you are right, but again, using Microsoft's main services I honestly have reasonable doubt that their servers would be the one's to get compromised, they are pretty much the most up-to-date company in terms of security practices, same can be said for Google also to a lesser extent.

@Azxdreuwa: The price to get around that (i.e. maintain another password) is imho much lower than this happening, but ymmv. There is a pretty nice black market for 0day exploits...

@PlasmaHH Oh I definitely agree with you, always best to create new passwords and, in some cases, aliases for each service used, to at least ensure some level of security.

@Azxdreuwa, I'm... not so sure I'd trust Microsoft's security that far. In a past life, when working somewhere that was doing an experimental partnership with Bing, we had to rearchitect our system to route billing data outside of the data path forwarded through their website, since they didn't trust their web frontends to handle PII.

@Azxdreuwa Using all of Unicode for passwords is guaranteed to give you loads of problems; many sites already choke on a space or quote or backslash. But hey, let's run with it. 100 random Unicode 10.0 characters give you log2(136690^100) ≈ 1706 bits of entropy, so 1706 coins, still a far cry from 100 000. I agree that that what you propose is outside of the realm of guessability! I disagree that it is less likely than 100 000 coins landing heads. The math clearly shows that it is barely less likely than 1706 coins landing heads. 595 coins if sticking to [A-Za-z0-9] for max interopability.

@marcelm thanks a lot for all of the analysis you've done on the entropy of the password, it's been quite an eye opener, and has made me realise that I was way off with my 100k coins. Funnily enough, Reddit does support Unicode characters as a password, including spaces etc., and a very long password length.

@Azxdreuwa, ...my day job is making a product that wouldn't have any cause to exist if Windows were adequately secure. Can't go into details, but -- sorry, I'm not there with you.

Let's assume that MS is actually at the forefront of data security - what happens when someone finds a bug inside the Intel or AMD CPU that runs their servers that lets the attacker compromise the secure software? I'm not sure how you can be so confident given that you know they don't build and manage every single thing involved in running their servers.

@Azxdreuwa : Microsoft is indeed pretty security conscious - but they are also a very juicy target. I assume that the NSA just asks nicely if they want something, but the Chinese and the Russians will have to do it the hard way; GCHQ and BND will have to do it the hard way and use some sort of plausible deniability - however, I'm sure all of them do have some sort of access.

@iheanyi Such an attack vector affects everyone, and it would affect Microsoft in such a case true, but even then I am sure Microsoft has mitigation procedures in such a situation to minimise any damage that may be caused by such as soon as possible.

@MartinBonner Whatever the NSA have, the GCHQ have also, they do work extremely closely to each other, but then again they are government agencies not a malicious group, they're not going to do anything with that data except for analysis etc., unless your a target of sorts.

Ok, I suppose it's progress at least that you've learned password reuse is a bad idea. I'll leave the blind trust in any one entity thing alone.

Oh no I've know for a while that reusing passwords is very bad, I've been hit by a fair amount of data breaches across multiple websites, including a large trading one. I was more interested in the risk posed through password re-use of a very strong password, but in that case if you can generate a very good password, might as well use a password manager

I'm definitely not trusting of any company, but in the case of Microsoft, they are truly certifiable, their cryptography and authentication libraries speaks for itself, they definitely know what they are doing.