[Discuss] Steve Gibson's SQRL

[Richard Pieri]

On 2/25/2015 1:18 PM, Tom Metro wrote:
> also connect to the wrong end-point (though the attacker could proxy the
> connection).

Which is trivially easy to do when providing victims with malicious URLs 
via malicious QR codes.

> -the domain in the URL is shown to the user for verification before the
> key is sent to the authentication web service.

We browsers have had this capability since Day 2, maybe Day 3. It still 
is not a reliable deterrent to users clicking on malicious links.

> -there are optional modes where it may be required that the browser and
> the authentication app both connect from the same IP.

IP address spoofing is easy.

> -the communication channel with the authentication app serves as an "out
> of band" channel that can optionally be used to verify high risk
> operations, such as when making a transfer between banks.

It's not out-of-band. The attacker can route both authentication and 
live sessions through his proxy. That's in-band.

> If the attacker has the ability to alter the DNS responses the user
> sees, and forge SSL certs for the site and authentication service, then
> they still might be able to pull off a MitM attack.

Altering DNS responses is unnecessary when the attacker can route the 
victim's traffic through his proxy (see previous). Likewise forging SSL 
certificates when he can run his own CA.

I like APOP. It's a robust idea. I do not like QR codes for encoding 
URLs or authentication information. Fraudulent codes are effectively 
impossible for human beings to identify. Trusting unverifiable blobs for 
security does not seem like a good idea to me.

-- 
Rich P.