AI Quote: Bitchat uses Noise Protocol encryption, but it does not automatically eliminate a man-in-the-middle attack during first contact.

For private Bluetooth conversations, it uses:

`Noise_XX_25519_ChaChaPoly_SHA256`

The Noise XX handshake:
Generates temporary session keys.
Exchanges each device’s persistent public identity key. Derives separate encrypted transport keys.
Uses ChaCha20-Poly1305 to detect modification and encrypt messages.
* Provides forward secrecy, so later theft of a long-term identity key should not expose previous session traffic. ([GitHub][1])

This prevents ordinary mesh relay devices from reading or silently modifying private messages. Intermediate phones forward encrypted packets; they do not possess the conversation’s end-to-end session keys. ([GitHub][1])

## The important limitation

Noise XX authenticates cryptographic keys, but on first contact it does not inherently prove that a key belongs to the person whose nickname you see.

An active attacker positioned between Alice and Bob could theoretically:

1. Establish one encrypted session with Alice.
2. Establish a separate encrypted session with Bob.
3. Present their own key to each side.
4. Decrypt and re-encrypt messages between them.

Both connections would be encrypted, but each would terminate at the attacker.

Bitchat addresses this through fingerprint verification. Every identity has a fingerprint calculated from its long-term Curve25519 public key:

`SHA-256(Noise static public key)`

Users must compare that fingerprint through a separate trusted method—ideally in person, by QR code, over a trusted call, or through another authenticated channel—and then mark the contact as verified. ([GitHub][1])

Once you have verified and saved the correct fingerprint, an attacker presenting a different identity key can be detected. The project’s own whitepaper explicitly says that users must verify fingerprints out-of-band to prevent MITM attacks. ([GitHub][1])

So the accurate description is:

> Bitchat prevents passive interception and protects messages from mesh relays automatically. Full protection against an active first-contact MITM requires users to compare and verify identity fingerprints.

Simply recognizing the username is not enough because nicknames are not inherently trustworthy or unique.

[1]: https://github.com/permissionlesstech/bitchat/blob/main/WHITEPAPER.md?utm_source=chatgpt.com "WHITEPAPER.md - permissionlesstech/bitchat"