Why Seconds Mean Hundreds of Millions
Two transactions, 115 seconds apart. Valid-looking signatures, no matching withdrawal request.
382 attack transactions over 152 minutes. One message replayed by wallet after wallet.
Research shows the opening burst was detectable inside the first ten minutes. That is the bar BridgeSentry is built to clear.
Burn Through the Exploit Payload Veil
Anatomy of $2.8B in Cross-Chain Breaches
Historical exploits broke deterministic state invariants. BridgeSentry mathematically correlates deposits, proof roots, and relayer timings to halt unbacked payouts in real time.
Nomad Replay Swarm
A routine contract update marked the 0x00 empty root as valid. Copycats cloned payload calldata and drained $146.6M across 382 transactions in 152 minutes.
Correlator detected payouts with zero deposit root on source chain. Flagged by Invariant R1 (Orphan Release) and Behaviour R6 (Swarm Velocity).
Noisy-OR Aggregation & Whale Guard
Dual invariant gating ensures that legitimate high-volume whale transfers never trigger false-positive alarms, while zero-root forged mints are flagged with P ≥ 0.999 certainty.
P(Attack) = 1 − ∏ (1 − pᵢ)Independent weak anomaly signals compound naturally into high suspicion. No single noisy metric can trigger a false positive on its own.
CRITICAL: inv ≥ 0.8 & usd ≥ $1MLegitimate large-volume transfers move millions safely. High volume alone never triggers CRITICAL severity unless a protocol state invariant breaks.
risk(t) = risk(t₀) · e^(−λ Δt)Adversary wallet risk scores surge instantaneously on malicious execution and cool down exponentially, preserving memory across multi-step drain cycles.
Detection Rules Catalog (R1 – R10)
Target payout has no corresponding origin deposit (ORPHAN_RELEASE).
Target withdrawal amount differs from source lock amount beyond slippage.
Signatures outside validator set or ≥80% quorum monopolized by one operator.
10-minute outflow z-score ≥ 3 against 6-hour moving baseline.
Single withdrawal draining an anomalous percentage of pool TVL.
Cluster of distinct beneficiaries replaying similar payload signatures.
Payout executed before origin transaction confirmed consensus finality.
Identical cross-chain message ID processed more than once.
High-value payout directed to wallet with zero bridge history.
Execution nonce or message sequence jumping outside sequential bounds.