[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

[tor-dev] ScrambNet: anonymity with an information-theoretic content floor (pre-alpha, seeking attacks)



Hello tor-dev,

Apologies for the cross-project note; this list is where path selection,
post-quantum, and attack analysis actually get discussed, which is why I'm
bringing it here rather than anywhere louder.

ScrambNet is a pre-alpha anonymity-network design: a threat model, a growing
RFC-style wire specification, and a runnable proof-of-concept. I'm posting it
to be attacked while nobody relies on it yet.

The one load-bearing idea, and the only thing that makes it different from Tor:
message *content* is protected information-theoretically, not computationally.
Each message is split with Shamir secret sharing into n shares (strawman
n=4, k=3); any k-1 shares are uniformly independent of the content, so there is
no ciphertext to break below the threshold -- today or against a future quantum
adversary. Shares travel over disjoint circuits to dispersed dead drops and the
recipient collects any k. Sphinx/onion crypto is kept, demoted to a second
belt: if it ever falls, the network does not become readable, only less
convenient.

This is only a claim about content. Linkage/metadata is the same hard problem
everyone else faces, and the repo says so plainly.

What the repo is deliberately honest about:
- the content floor costs x n bandwidth (x4 at the strawman);
- fast mode does not defend against a global passive adversary, as Tor
  declares for itself;
- dead-drop retrieval leaks polling cadence to a drop within an epoch;
- a small network cannot build disjoint circuits and must degrade *and say so*,
  never silently.

The load-bearing parts are executable (pure-Python standard library, no
dependencies): k-of-n dispersal with a brute-force demonstration of the
unconditional floor, dead-drop delivery with disciplined polling, and a k-path
adversary simulator that reports content-break and linkage-leak rates against
network size. Writing the code already forced two specification fixes.

The work most relevant to this list, and where I'd most value a hostile read,
is path selection: verifying that n circuits are genuinely disjoint (node, AS,
jurisdiction) using only a client's own gossip sample, against relays that lie
about everything except their signatures. It is drafted (spec Part 3) and
openly unsolved (issue OP-1).

Open problems are filed as numbered issues (OP-1..OP-9), each asking for
attacks with numbers rather than opinions. There is a comparison table against
Tor, Nym/Loopix and Briar, including where ScrambNet is behind.

  Spec, PoC, issues: https://github.com/ScrambNet/scrambnet

Everything is CC0. Pseudonymous contributions are the norm there.

If the core wager is wrong, the fastest way to find out is to break it, and
I'd rather that happen here than in deployment. Thanks for reading.

_______________________________________________
tor-dev mailing list -- tor-dev@xxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to tor-dev-leave@xxxxxxxxxxxxxxxxxxxx