Security model
This page sets out what QUAC protects against, what it relies on, and where the edges are. To report a vulnerability, see SECURITY.md. Please report privately, not in a public issue.
Threats QUAC is designed for
A large quantum computer
Goal: an attacker with a cryptographically relevant quantum computer can't steal funds or forge blocks.
| Asset | Protection |
|---|---|
| Account funds | Spending needs an ML-DSA-65 signature. ML-DSA-65 is designed to resist quantum attacks (NIST category 3). Classical keys can't sign at all: the ML-DSA-only check rejects them before execution. |
| Block finality | Validator votes are ML-DSA-65 signatures. Forging a commit needs forged ML-DSA-65 signatures from more than ⅔ of voting power. |
| Addresses | SHA-256 based, so Grover's algorithm gives at most a quadratic speed-up. Exposed public keys are ML-DSA keys, which Shor's algorithm doesn't help with. |
"Harvest now, decrypt later"
Blockchain data is public by design, so there is little confidential data to harvest. Signatures are what matter, and QUAC's are post-quantum from genesis. Recorded P2P traffic, encrypted with X25519 today, would reveal only gossip that was public anyway.
Classical attacks
QUAC inherits CometBFT's Byzantine fault tolerance: safety holds while validators with less than ⅓ of voting power are malicious. Slashing penalizes double-signing (5% of stake by default) and long downtime.
Assumptions
- ML-DSA-65 is secure. QUAC relies on FIPS 204 and on the implementation in CometBFT/Cloudflare CIRCL. A break in ML-DSA would affect QUAC the way a break in ECDSA would affect Bitcoin. The chain can move to a new scheme by software upgrade, but the network has to act before any break is exploited.
- Validator keys are kept safe. A stolen consensus key can double-sign, which gets the validator slashed. A stolen operator key can move the validator's funds.
- Governance acts in good faith. See below.
Known limitations
Classical P2P transport
CometBFT nodes authenticate each other with Ed25519 node keys and agree on session keys with X25519. A quantum attacker could impersonate a node at the network layer. That would let them eclipse or censor a node, but not forge blocks or transactions, which carry ML-DSA signatures checked by every node. A post-quantum P2P handshake is on the roadmap.
Governance
Governance is the chain's upgrade mechanism, so it can in principle undo QUAC's guarantees:
- a consensus-parameter proposal could add
ed25519back tovalidator.pub_key_types, - a software upgrade could remove the ML-DSA-only check.
Either would need a governance vote by stakers and, for an upgrade, validators running the new software. QUAC's position is that classical signatures will never be re-enabled. Treat any proposal that does so as an attack.
Funds sent to classical addresses
The chain can't tell what kind of key an address came from until the key signs, so funds can be sent to an address derived from a secp256k1 key, and those funds can never be spent. Wallets should only produce ML-DSA-65 addresses. The explorer flags accounts holding classical keys.
Signature size and block space
ML-DSA-65 signatures are about 50 times larger than Ed25519's. Each validator adds 3,309 bytes to every block, and every transfer is about 3.6–5.5 KB. Size-based gas (tx_size_cost_per_byte) prices this in, and block.max_bytes and block.max_gas bound it. A large validator set needs a careful look at bandwidth and storage.
Testnet specifics
- Testnet keys were generated on one machine during the genesis ceremony and are stored with the
testkeyring backend. That's acceptable for a valueless network and must not be repeated for mainnet. - Four equal validators tolerate one validator down, but two of them (
quac-val-1,quac-val-3) share a server, so losing that server pauses the chain. - The faucet's rate limits (per address and per IP) slow down hoarding but can't stop a determined user with many IPs. Testnet QUAC has no value, so this is accepted.
Operational security
- No published infrastructure IPs. Public endpoints go through a Cloudflare Tunnel. CometBFT RPC, whose
net_infolists peer IPs, is not published. Genesis gentx memos carry placeholder addresses. See Public testnet. - Least privilege. Nodes run as an unprivileged
quacuser, and the node home is mode 700. The REST API and the faucet listen on localhost only. - Firewalled P2P. Port 26656 is open only to known peers.
- No secrets in the repository. Generated keys go to
deploy/testnet/out/, which is gitignored.
Upstream code
Most of QUAC's code is the Cosmos SDK and CometBFT, and QUAC tracks their security releases. QUAC's own consensus-relevant code is small and covered by tests:
app/pqante.go(about 115 lines)app/tokenomics.goapp/config.go