From f271d5921a7475887a6c9bc1e7068c4b1ad672af Mon Sep 17 00:00:00 2001 From: sonix aka privex Date: Sun, 27 Sep 2026 19:07:19 +0530 Subject: [PATCH 1/2] docs: name two disruption limits in the security design - What Privex doesn't protect against: whoever holds a session token, the TLS-terminating proxy, or a compromised server can make new contacts' first messages fail (one-time prekeys are unsigned by design, as in Signal) or empty the message queue - never read anything. - Known limitations: one mailbox per account - with two active devices, the one that can't decrypt a message confirms it and the other never gets it; per-device mailboxes would reveal your device count, so they aren't built. Co-Authored-By: Claude Opus 5.5 --- docs/SECURITY_DESIGN.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/docs/SECURITY_DESIGN.md b/docs/SECURITY_DESIGN.md index cdf6f2e..99d607b 100644 --- a/docs/SECURITY_DESIGN.md +++ b/docs/SECURITY_DESIGN.md @@ -120,6 +120,7 @@ A security tool that won't name its limits shouldn't be trusted. Privex does **n - **Voluntary exposure.** If you share your identity or messages, that's outside the model. - **Legal risk in your own jurisdiction.** Privex provides *technical* privacy, not legal immunity. It can make your data impossible to produce; it can't make an activity legal where you are. - **Traffic-analysis perfection.** Mixnet delays plus cover traffic narrow the correlation window dramatically but don't provably close it to zero. This is an acknowledged, industry-wide hard limit, not a solved problem. +- **Disruption by whoever holds a session token or sits in the path.** Someone with a valid session token (tokens last up to 24 hours), the TLS-terminating proxy in front of the server, or a compromised server can make new contacts' first messages fail by swapping your one-time prekeys (unsigned by design, as in Signal's protocol), or empty your message queue by confirming its messages. They can't read anything, and the same parties could simply drop your traffic. --- @@ -132,6 +133,7 @@ Being specific here is the point of open-sourcing before it's finished. - **Solo-maintained.** One person's blind spots are real. Independent review is the mitigation, and it's wanted. - **File sharing is disabled** in Phase 1 pending client-side, zero-knowledge CSAM protection (PDQ perceptual hashing + OPRF-based PSI + a Groth16 proof the server can verify without seeing the file or the hash). A messenger this private has a duty not to become a safe distribution channel for abuse material; shipping that safeguard is a precondition for enabling files. - **Groups and calls** (MLS, WebRTC + SFrame) are specified and partly scaffolded, not shipped. +- **One mailbox per account.** Live messages go to your newest open connection. With two active devices, one that can't decrypt a message (its chat session lives on the other device) confirms it, and the other never gets it. Per-device mailboxes would fix this but would tell the server how many devices you have, so they aren't built. --- From b3b3347882192e02a45f3350e334ae434969eda5 Mon Sep 17 00:00:00 2001 From: sonix aka privex Date: Sun, 27 Sep 2026 19:15:46 +0530 Subject: [PATCH 2/2] docs: tighten the disruption-limit wording - A token holder isn't in the path, so only the proxy or server can drop traffic. - Signal signs its one-time post-quantum prekeys; the comparison is to its one-time curve prekeys, which Privex's are. - Any new chat session draws a one-time prekey, not just a new contact's. Co-Authored-By: Claude Opus 5.5 --- docs/SECURITY_DESIGN.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/SECURITY_DESIGN.md b/docs/SECURITY_DESIGN.md index 99d607b..13510da 100644 --- a/docs/SECURITY_DESIGN.md +++ b/docs/SECURITY_DESIGN.md @@ -120,7 +120,7 @@ A security tool that won't name its limits shouldn't be trusted. Privex does **n - **Voluntary exposure.** If you share your identity or messages, that's outside the model. - **Legal risk in your own jurisdiction.** Privex provides *technical* privacy, not legal immunity. It can make your data impossible to produce; it can't make an activity legal where you are. - **Traffic-analysis perfection.** Mixnet delays plus cover traffic narrow the correlation window dramatically but don't provably close it to zero. This is an acknowledged, industry-wide hard limit, not a solved problem. -- **Disruption by whoever holds a session token or sits in the path.** Someone with a valid session token (tokens last up to 24 hours), the TLS-terminating proxy in front of the server, or a compromised server can make new contacts' first messages fail by swapping your one-time prekeys (unsigned by design, as in Signal's protocol), or empty your message queue by confirming its messages. They can't read anything, and the same parties could simply drop your traffic. +- **Disruption by whoever holds a session token or sits in the path.** Someone with a valid session token (tokens last up to 24 hours), the TLS-terminating proxy in front of the server, or a compromised server can make the first message of any new chat with you fail by adding or replacing your one-time prekeys (unsigned by design, like Signal's one-time curve prekeys), or empty your message queue by confirming its messages. They can't read anything, and the proxy or server could just drop your traffic anyway. ---