Key Exchange Problems - Computerphile 

"Oh yes, I'm Bob!" ... _But he isn't_

I love how he puts so much effort into the diagrams, and then they just make a digital animation for each diagram anyway

RSA key exchange is fully susceptible by man in the middle as well. Sean has a private key and signs the data with its own key. Alice cannot know wether the data was encrypted and signed by Sean (man in the middle) or Bob since the real identity cannot be verified by an external authority. This is why we have certificates. SSL is what websites (HTTPS) are using to implement the whole chain of security features to finally become secure against man in the middle attacks. The final step that SSL does ontop of RSA key exchange is to verify the public key with the certificate that server sent to the client upon SSL negotiation, using a global certificate store. In short, certificates themself have to be signed by a certificate authority, which (typically) can only be modified by Windows Updates (for Windows) and alike. The certificates for HTTPS include the domain name in it's ServerName property to restrict the usage of the certificate to a particular website. The browser will make sure to verify this. I think this should have been mentioned in the video before people run off and use RSA by itself when it really isn't secure against man in the middle (but it is secure against capturing of the data where it is not re-encrypted). Side note, SSL includes all above mentioned features (configurable). If you're interested in playing around with this in programming, try out the OpenSSL library. Also HTTPS is typically using an SSL library such as OpenSSL. For example, Chrome uses "boringssl" which is a library 'forked' from (based on) OpenSSL.

The man, the legend, Dr. Mike Pound!

Please keep doing this!

that thumbnail face tho

I love how the host is so attentive and asked the question at the end. I had the same while watching the video

* waiting for the elliptic curve cryptography video impatiently *

The best part of this video is when the interviewer says, "Diffie Hellman is dead in the water" and Dr. Mike Pound(with the most hilarious expression says, "Diffie Hellman is in REAL TROUBLE HERE!" I couldn't stop laughing and laughing! Awesome video.

Mike Pound is love. Mike Pound is life.

Mike is absolutely phenomenal! Rarely you see someone so knowledgeable and soooo funny at the same time. To use one of his words - "brilliant"!! :-)

Forgot to mention the necessity of being able to safely share the public key, otherwise Sean could just nab that as well and do the same attack (why yes, I am Bob! You can verify that by checking me against the public key I just sent you!) That's where things like certificate authorities come in -- a (hopefully) trusted third party that can retain Bob's public key for him such that he doesn't have to send it to Alice himself and therefore Sean has no chance to inject himself into the conversation. Of course, that just punts the problem up a level: How can you trust that Bob's public key actually came from the CA? If Sean is operating at Alice's end of the connection, he could potentially intercept communication to the CA server as well as to Bob. As far as I know (and I might not be entirely accurate here..) this is resolved primarily by your OS and/or browser having a built-in list of trusted CAs (and we just assume that Sean hasn't been able to hack her browser or OS install.. if he had that level of access to Alice's machine, the whole question is moot anyway since he could just install a keylogger or whatever and capture the session directly.) So the CA send Bob's public key and authenticates themselves using their private key.. Alice can then use the CA's public key that she has stored locally to verify them, allowing her to safely retrieve Bob's public key and in turn use that to verify Bob. But that means trusting the CA (in the social sense, rather than the computational sense.) There was one big one from China this past year.. maybe 2..? that Google removed from Chrome's trusted list and the other major browsers slowly followed suit, because the CA wasn't acting trustworthy and could have potentially compromised security by double-issuing certificates and back-dating expiry dates and things like that. For the most part though, that's not a huge problem since CAs are basically out of business when the browsers stop trusting them -- meaning they have a huge incentive to play by the rules and those that don't won't matter for long either way.

I discovered this channel a few days ago and I am already addicted to it... I have watched a dozen of your videos already and I liked each and every single one. All the computerphile team has a gift, I wish I could explain things the way you do. One question came to my mind while watching this particular video though: After having watched this video it is clear why, even with the server's public key, an extra "final" key is needed (the one that is going to be used to encrypt the requests and the requests and the responses before the transmission) using the DH key "exchange" (I used the quotes because in another video it is very well explained that this key is not really exchanged but generated at both sides equally instead). But the generation of this key has a computing cost at both sides, especially at the server side which needs to generate DH symmetric keys for all clients that are connected to it. My question then is: wouldn't it be better that, instead of using DH, the client generated not a random symmetric key - not a good idea in case the server certificate's private key gets compromised, as it is explained in this video - but a random pair of private/public asymmetric keys and sent the public one to the server? (just keys, not certificates, no need to check any certification validity at this level), so the protocol would look something like 1) client -> hello -> server 2) server -> certificate with public key -> client 3) client - checks the certification's validity and, if valid, generates a random pair of private/public keys 4) client -> client's public key from step 3 encrypted with the server's public key from step 2 -> server 5) server -> confirmation message encrypted with the server's private key (authenticity layer) + client's public key (confidentiality layer) -> client 6) client - checks that the message from step 5 can be decrypted (this would confirm that its public key has really reached the server). 7) client -> data request, encrypted with client's private key in 1st place (authenticity) + server's public key (confidentiality) -> server 8) server -> requested data, encrypted with server's private key (auth.) and client's public one (conf.) -> client The benefit I see in this approach is that the cost of generating "final" asymmetric keys, even though it is going to be probably higher than using the DH process (maybe even higher than twice - the sum of the server's and the client's costs in DH), relies completely on the clients and this would give some rest to the server, which is the one that suffers the most. In case these keys generation part would take "too long" for the client, the keys generation could start asynchronously at the same time that the hello message is sent at step 1, and step 4 would have to wait until the client's keys are completed; of course, only if the server's certificate has been verified... but even if the server's certificate is not valid, that pair of keys could still be used/recycled by the client in a different session/web (even with a different server), so their generation would not be a waste of computing power, and would make the process faster for that other session - this would need a thorough testing by the client's developer to make sure that a pair of random keys is not erroneously flagged as "still-usable" (once assigned to a session, they should NOT be used by another one). Maybe this is all madness... but your videos made my imagination fly 😅

thank you for uploading and have a happy new year. cheerio Toni. PS: I really like all of your films, they are totally informative for me, cheers

I am taking network security class in college and this video explore a little more in depth of what I have learned so far. Very satisfied all the works from computerphile. :)

Dr. Conspicuously Inconspicuous Smirk is back!

@2:10 "He isn't" That look though...XD

6:00 “Other nefarious people are available” 😂

I paid £9000 to learn this

Great video. Understanding these sort of key exchanges and realising how they can be broken by a Man in the Middle attack like this shows just what a huge security problem Superfish was (and probably still is on some computers). If you haven't seen it, look for the Computerphile video with Tom Scott from 2015 called "Man in the Middle Attacks & Superfish".

This series of videos are the best explanations ever thank you so much

I’ve watched most of this stuff on encryption and I don’t fully understand it, but this chap is brilliant at explaining what is going on plus the pros/cons of each system. Engrossing.

But then, how can we verify that the public key is actually Bobs public key? *Insert root certificate explanation here

THANK YOU - this is the info I've been looking for everywhere!

I didn't understand much of this, but I love listening to Dr. Pound.

Not only does RSA hope that your private key doesn't get leaked, it also needs to assume that only Bob can get an RSA key pair for his domain name. Anyways, great video guys!

Aaaah, the new pens are glorious. Thank you.

Finally, a computerphile video I actually understand, thanks to my knowledge with PGP

OK, so... We use RSA to verify your identity versus a database of public keys, in order use Diffie-Hellman to make a short-lived shared key, all to send a one-time private key for use with AES for the final communication. I love it!

I have been wondering for a long time why we can't just encrypt the symmetric key with RSA and now i finally know the answer!!!!! Thank you :)

Holy moly Dr. Pound is amazingly charming! Especially so when discussing nefarious business! :D

This video is really great. I'm glad I found it

Strictly speaking, it's not RSA that rescues it, but A's existing knowledge of B's public key. Otherwise, the network can step in and say "I''m the server you requested, and this is my public key," and you have gotten nowhere. Usually, it is some certificate authority that is built into the browser. But the point is there needs to be a public key that the man in the middle can't lie about.

Thank you for answering my question!

lucid and succinct! nicely done!

Nicely explained, thanks!

Hi, i love ur videos :) Could you do Kerberos including weaknesses?

This is awesome. Best explanation ever

Mike Pound fan here, keep it up! :)

we need a video on the Intel #Meltdown bug pronto

Clear explanation, I am master student at Turkey, security course lecturer doesn’t provide this explanatory. Thanks

Loved this video!

These explanation Videos are superb! Now i have some hope, that i can get through my exams :D

my professor actually linked this video in the homework and had us watch it and the homework had questions about it

Best. Thumbnail. Ever.

This is beautiful.

I'm glad Dr Pound listens to the 2 serving suggestion on the Pepsi haha

excellent video, are you going to do one on public key infrastructure to show how certificates are trusted, and how it ties into RSA?

Eye opener. I was going to use ECDH. Thankfully now I know I should youe ECDH+RSA.

This video sidesteps a very important problem. How do I know I have Bob’s public key? For example, when Alice asks for Bob’s key, Sean can intercept that and send his own key, masquerading as Bob. You only pushed the problem one step down the stairs.

Meltdown and Spectre soon pls

The "He isn't." face

Hi, first of all thanks for these videos about encryption. As a programmer I often hear that I shouldn't create my own encryption protocol and use something like TLS, now I see why :D so after seeing this I have one question. How can one establish safe connection in decentralized p2p network where is no CA for RSA key-pair creation? It would be nice to have a video on this topic :)

I’m starting with the man in the middle I’m asking him to change his ways And no message could have been any clearer If you wanna make the world a better place Take a look at yourself, and then make a change

We need a video on speculative execution!

Can someone please explain how does a key exchange take place in case of DES / AES algorithm or in that matter any symmetric key algorithm? By the way excellent videos!!

thank you very much the video is really helpful

Awesome video!!! Thanks

Do the Spectre CPU flaw!

Thanks for spelling "Sean" correctly.

I have a question ..... is 'g' same in both cases (Alice and Bob) or it can be different for both. How is 'g' and 'n' selected/shared before everything?

If there are no certificates involved (like SSH), this will only work if the first exchange is safe, right? From what I understood, if the first exchange is compromised then the following ones will be too. Can someone please give me some insight?

Is the ephemeral key deleted from the server once its use is over? Like every day or so? (Or per session) If it is stored, it leads back to the same problem of using RSA key. If someone could break in to the server, they have got the keys (RSA AND ephemeral) and if they had stored the encrypted communication between client and server, security breaks down.

1:25 Not calling the attacker Eve. Instant dislike, unfollow and uninstall.

Please talk about Post-Quantum Cryptography!

Shaun? What happened to Eve? :)

I'm just glad you people are thinking hard about this, whilst I do my on-line banking on faith alone (so far).

the scenario he explains is when the bob and alice already trust each other (they communicated before and alice knows bobs public key), but what about when a connection is being established for the first time (bob has to give his public key to alice)? is man in middle possible still?

That is a genuinely terrifying thumbnail.

beautiful

What a legend!

Question.....what would likely cause a Key exchange server password corruption? Is it likely to inexplicably fail without any type of human interface if it is encrypted?

I have a stupid question.. so bare with me please You sayed that the server bob sign his 'gb' calculation with his private key. Does that mean he encrypt it with his private key just like what the public key can do ?! Im confused

been wearing that same sweater in the last 3 videos bro!

If I already have a pre-shared public key of bob and know for sure that it belongs to the private key of bob. Is DH enough to authenticate the server? I think the key exchange will always fail to produce the same value when there is someone else pretending to be bob. Right?

I genuinely don't understand why use RSA public keys (which themselves have to be certified by a central authentication service) if we could just use a three-pass Diffie-Hellman without the need of a centralized authentication service to certify the RSA public keys in the first place. I'm pretty sure that without the central authentication a man in the middle can do the same thing to RSA public keys as they can with single-pass Diffie-Hellman. But if you could share ephemeral session secrets with 3-pass Diffie-Hellman why not do that. Those ephemeral could be the temporary RSA public keys (or anything else really) right? Edit: This is a legitimate question. I am taking an introductory infoSec undergrad course and I want to make sure that I understand DH and RSA correctly before the coming midterm. I appreciate any clarifications

So after they establish the shared key to encrypt and decrypt finally, after all this, how long does it last? Is it terminated after a session is finished? Saved in a cookie? Or, for every minute a new shared final key is created going through all this RSA process again? Or does it depend upon us (client,server) to decide how long it lasts and there is a default time set which we can change.

The digital signature doesn't stop the man-in-the-middle attack because the attacker can gain access to the public key for decryption. I mean both Alice and Bob can realize that their communication have been intercepted but if Bob in that scenario sends something important right from the very start, that's basically it

Did you cover how the public key is distributed in the first place? How do we verify a public is not one just sent from Sean?

I read somewhere long ago that not only was bulk encryption of data using RSA inefficient, but that there was issues with encrypting anything longer than the private key. I can't find where I read this, but could someone please explain this?

5:38 How did Alice knew what process had to be done with bg to get the same encrypted message which Bob made with his bg? Anyone help!

I didn`t get it. What is preventing Sean from acting exactly like Bob in the second scheme?

i think it needs more details? because people will assume that as long as you are the man in the middle at the start! you can still decrypt and get all the messages and just pretend to be bob, you couldnt jump in half way through for sure with this explanation, but you forgot to mention that certificate authorities give pre installed keys to peoples computers so that the initial handshake doesnt get hijacked. that is explained in a previous video as i remember

am i missing something? , for me RSA and DH(diffie hellman) could be used for the same thing ,why cant we share g to the b as a public key on the certificate, the problem of authentication is a totally different problem and these two algorithms both establish a secure channel , to know the other end u need certificate in a tree signed by root certificate that is embedded in browser or OS , also we could use RSA two time upon each other each time generating a new inner key pair, (i assume maybe DH is more performant and RSA more secure is the only cause ?, if so it should have been mentioned as such ,if not could someone correct me?)

Make a video about meltdown/spectre please!

But does that not assume, that A knows B’s public key? So if it is not shared on a piece of paper, can’t Sean just MITM the public key exchange? Is there any way of fixing that without something like SSL/TLS?

How do you derive the asymmetric keys? More maths please!

How RSA is preventing from MITM? MITM is not only about changing a message but also about reading it. So in presented scenario Shaun can still reads what Bob is sending to Alice and vice versa, as Bob in only signing a message, not encrypting it.

What's to stop Sean doing the same man-in the middle attack on the private/public keys? It would require an out-of-band secure communication for Alice and Bob to know the true public key of each other.

Normally as a client you'd have access to your server's public key, and Alice will send her message encrypted with the server public key so only your server will be able to decryped it, and same on the way back. When it comes to other services, you have a certificate authorities to verify the server Alice is talking to is really who it claims to be.

This breaks if Sean is making modification while Alice and Bob are exchanging public keys. Sean can listen to Alice sending her public key (AlicePub) to Bob and instead of letting that public key reach Bob, he can send his own public key (SeanPub1) to Bob. Now Bob will sign the message containing Bob's public key (BobPub) using what he falsely believes to be Alice's public key (SeanPub1) and try sending it to Alice. Sean, man in the middle, then can decrypt the message and send another public key (SeanPub2) to Alice. In this way Bob successfully deceived Alice and Bob about each other's public keys.

Let's just say that we wanna build a multi-client chat app. Does this still work in that case?

As far as I understand Server sends the G, N and G^random signed with servers private key and sends along with the server certificate to client, once the client verifies the authenticity using PKI and he will send his G^random and encrypts this using servers public key. Thus they will both arrive at the shared secret key using DH with RSA. These keys are ephemeral, i.e a new key pair is generated for every new session with the server.

5:15, still the man in the middle can read the data right ? (not modify, but still read it)

How does using RSA prevent Sean from spoofing Alice's request to get Bob's public key. Couldn't he not block that and send his own?

One other problem that we would face if we only use RSA for sharing secret key would be that we would need public key of Alice (client) to encrypt the secret key. And in almost all the cases of web traffic, only servers have digital certificates, not the client.

thank you so much sir

How and when did Alice get Bob's Public key, for verification?

how difficult is it to generate public private keys for RSA. is it something that i could do on my home computer or do i need to trust a key generating service not to keep a copy of the private key they made for me.

Now, since the basics are done, the only thing left is to talk about the SSL and Certiciface Authories and the delegation chain, right?