This policy covers this repository: the Rust serverless node workspace under node/, the protocol specifications under protocol/, the Python reference implementation under protocol/reference/, the shared test vectors under shared-vectors/, and the scripts / CI under scripts/, node/scripts/ and .github/. Mobile apps and any legacy server live in other trees and are out of scope here (report issues in them the same way; we will route them).

• Email: info@raven-messenger.com
• Subject: [SECURITY] <short description>
• Do NOT open a public GitHub issue, pull request or discussion for a vulnerability.

Please include the affected component (crate / binary / spec / script), the commit you tested, build features used (for example whether unsafe-demo-crypto or a lab feature was enabled), reproduction steps or a proof-of-concept input, and the impact you believe it has.

In scope — default builds and anything reachable by a remote peer or another local user:

• Wire decoders and protocol logic in raven-core: RVN1 envelope (RAVEN_ENVELOPE_V1), ACK, store objects (RSO1), PairInit / pair response, ATSAM indexed sessions and RVNA1 bodies, prekey bundles and lifecycle, device certificates and revocation, routing / store tags, alias and profile records, contact-request / introduction codecs, RLB1 LAN offers, the LAN Noise binding (Noise_XX_25519_ChaChaPoly_BLAKE2s), bridge frames and store-and-forward queues.
• raven-node daemon: LAN / bridge / mock-BLE listeners, the Internet carrier, the local IPC endpoint (Unix socket / Windows named pipe) and its request parser, on-disk state (queues, session stores, chat history).
• raven-swarm: the libp2p host (Noise, TCP/QUIC, Kademlia /raven/kad/1.0.0, relay, DCUtR), signed PeerRecords and bootstrap config.
• ash CLI: identity creation and key storage (macOS Keychain, Windows DPAPI, Linux Secret Service), contact pinning, import/export paths.
• FFI crates (raven-fb-ffi, raven-mlkem768-incremental-ffi) and other unsafe code.
• Denial of service and resource exhaustion of peers, relays, bridges and store-and-forward nodes (unbounded allocation, queue or disk growth, amplification, CPU exhaustion from crafted input). Forwarding for others is a core function of a mesh node, so remote DoS is in scope.
• Specifications, reference code and vectors that are wrong in a way that would lead a conforming implementation to be insecure.
• Build, release and CI: anything that lets unsafe-demo-crypto or a lab feature reach a release artifact, or that compromises the supply chain.

Out of scope:

• Lab-only paths used as documented: the unsafe-demo-crypto feature and --body-mode unsafe-interim (a public-key-derived demo cipher that is known to be readable by any observer), RAVEN_LAB_TEST_A, the locked-file debug key backend, and features marked experimental / lab. A way to reach them from a release build is in scope.
• Social engineering, and physical attacks on an unlocked device.
• Volumetric network flooding that no protocol change can mitigate.
• Vulnerabilities in third-party dependencies with no Raven-specific impact (please report upstream; tell us if Raven is affected).

Read these before assessing an issue — several limitations are already known and documented:

• protocol/SECURITY_ERRATA_RVN1_2026-08-13.md — normative errata and production hold for RVN1.
• docs/THREAT_MODEL.md and docs/crypto/ATSAM_THREAT_ASSUMPTIONS_V1.md.
• protocol/SPEC.md — which wire formats are frozen and which are implementation-defined.
• docs/EXTERNAL_REVIEW_PACKET.md — review scope and known gaps.

There is no central message server and no TLS server endpoint in this repository. Message routing is store-carry-forward over LAN, bridge, mock-BLE, mailbox and libp2p carriers.

We credit researchers who report valid vulnerabilities, with their permission.