[Cryptography] Licensing of cryptographic services in France

[Christian Huitema]

On 8/29/2024 11:59 AM, Phillip Hallam-Baker wrote:
> On Wed, Aug 28, 2024 at 11:12 PM Christian Huitema<huitema at huitema.net>
> wrote:
>
>> On 8/28/2024 4:57 PM, Phillip Hallam-Baker wrote:
>>> Steganography just means a slight reduction in payload capacity.
>>> instead of passing 1200 bytes of payload at a time, it would be maybe
>>> 1140. So worst case 5%-10% reduction in bandwidth.
>> That's not the figures that I am used to. The most common algorithm is
>> to encode the hidden message in the least significant bits of an image
>> encoding, but then the signal is occupying at most 10% of the bits. But
>> this is kind of an upper bound. If I understand correctly , there is a
>> whole body of research developing better steganography, and developing
>> algorithms capable of detecting the presence of embedded messages.
>> Everything else being equal, the higher the ratio of message to payload,
>> the easier it should be to detect the presence of steganography.
>>
> Steganography is simply concealing the fact there is a message there.
>
> In a world where HTTP crossed the net unencrypted by default, twiddling the
> LSB of images was the best we can do.
>
> In a world where everything goes over QUIC, we can disguise an encrypted
> two way MOQ session as a HTTP/3.0 over QUIC session.
>
> Steganography has always been double ended, creating easier camouflage to
> hide in is just as valid as making use of the existing cover.

Define "disguise". You can certainly have some externally visible 
metadata that says "this is a web session". In fact, MoQ runs over 
Webtransport, which itself runs over HTTP3, so technically MoQ is HTTP3, 
no disguise required. But the attackers and classifiers are not relying 
on the metadata. They are doing traffic fingerprinting, using machine 
learned algorithms that classify streams based on the patters and timing 
of packets, i.e., features that are visible through the encyption. 
Beating that is very hard, if even feasible. You could try to add enough 
fake traffic as chaff to the base application traffic and change the 
packet patterns, but AFAIK you have to use a very large amount of chaff 
to be effective, and that is by itself its own pattern.

-- Christian Huitema