Research
ASAS-BridgeAMM: Trust-Minimized Cross-Chain Bridge AMM with Failure Containment
Cross-chain bridges hold enormous value but fail in a binary way: once an attacker's proof or validator quorum is compromised, the entire locked reserve can be drained in minutes, with no…

- arXiv
- 2601.12434
- Published
- 2026-01-18
- Authors
- Shengwei You, Aditya Joshi, Andrey Kuehlkamp, Jarek Nabrzyski
AI summary
Overview
Research area: Cross-chain blockchain interoperability and Decentralized Finance (DeFi) security, specifically the design of bridge-coupled Automated Market Makers (AMMs) that survive partial failures. Filed under cs.DC (Distributed, Parallel, and Cluster Computing).
Technical level: Advanced. The paper assumes familiarity with constant-product AMM mechanics (Uniswap-style x·y=k curves), Byzantine fault tolerance models, smart-contract state machines, and oracle design.
Scope: The paper specifies, formally analyzes, and empirically evaluates a single protocol — ASAS-BridgeAMM — that introduces a "Contained Degradation" operational state between full functionality and catastrophic failure, using cross-chain message latency as a priced risk signal.
What This Paper Is About
Cross-chain bridges hold enormous value but fail in a binary way: once an attacker's proof or validator quorum is compromised, the entire locked reserve can be drained in minutes, with no intermediate defensive state. The authors argue that this is fundamentally different from traditional finance, where circuit breakers and price limits contain losses rather than prevent volatility. Their goal is to build a bridge that treats uncertainty (message delay, price divergence) as a quantifiable execution risk and automatically degrades service — raising haircuts, doubling slippage bounds, tightening withdrawal limits, and eventually halting outflows — so that a total loss becomes a bounded "bad debt" event.
Key Contributions
-
A formal risk model. The authors define a Byzantine Relayer threat model (a relayer that can delay, reorder, censor, and front-run messages, and may collude with a superminority of validators, but cannot forge consensus signatures) and a Contained Degradation Invariant intended to bound economic loss even when message-passing layers are compromised.
-
Protocol architecture with a three-state machine. ASAS-BridgeAMM couples a source-chain Ingress vault, an off-chain Oracle/Relayer Hub, and a destination-chain AMM. The protocol is specified as an atomic state machine with three mutually exclusive states — Normal, Restricted, and Halted — with deterministic transition conditions.
-
Security proofs. The paper presents theorems for Safety (bad debt bounded per epoch by
h_max · C_totalin the main text), Liveness (honest transactions settle within a boundT_max), and Manipulation Resistance (adversarial profit bounded byδ · V_swap + s_max · V_pool), with proof sketches in the body and fuller proofs in the appendices. -
Empirical validation on a mainnet fork. The authors replay 547 days of historical market data and specific exploit scenarios, run 100,000 Monte Carlo iterations, test 10 attack vectors, and sweep 125 parameter configurations. A reference Solidity implementation of approximately 450 lines is described.
Main Findings
-
73% reduction in worst-case insolvency. Across the 547-day replay, ASAS bounded maximum bad debt to 0.19% of collateral per epoch versus unbounded exposure (reported as "5%+") in the baseline bridge, which the authors characterize as a 73% reduction in worst-case loss exposure.
-
Solvency held in every Monte Carlo run. All 100,000 iterations remained solvent (P = 1.0000), including all 50,066 runs in the 25–50% drawdown band and all 3,678 runs in the "Extreme" combined-stress bucket (40%+ drawdown, 30+ minute latency). The tightest reported 95% confidence interval was [0.9990, 1.0000].
-
Bad debt stayed under the 0.2% threshold. Mean per-epoch bad debt was 0.001305 (0.1305% of collateral), maximum observed was 0.001900 (0.1900%), and the 99th percentile was 0.001891 (0.1891%).
-
Volume retention exceeded the baseline (104.5% during stress). Overall ASAS volume was 61.39M against 58.46M for the baseline (105.0%); stress-period volume was 21.71M against 20.77M (104.5%); normal-period volume was 39.68M against 37.69M (105.3%). The authors attribute this counter-intuitive result to a "flight to safety" effect: LP capital stays sticky because of confidence in the floor mechanism, so liquidity is deeper than in the baseline during volatility.
-
Protective states activated frequently but not destructively. Over 547 days, Restricted Mode triggered 141 times and Halted triggered 4 times, while the protocol still processed 54,700 swaps.
-
The Orbit Chain replay produced 8 circuit breaker activations with 103.29% volume retention. Under the paper's counterfactual, the observed 15-minute latency would have raised the haircut from 0.3% to 2.65%, and extraction would be capped at 5% of the pool per epoch.
-
Defense-in-depth in the attack-vector sweep. Of 10 tested vectors, 8 triggered protection. Two (the gradual 25% drop and the sandwich attack) did not trigger the circuit breaker but still reverted on the slippage/solvency check. The maximum stress test (50% crash, 90-minute latency, 8 BTC swap representing 80% of the pool) resulted in zero bad debt.
-
Parameter sweep identified a Pareto-optimal setting. All configurations with
θ_price ≥ 50%achieved 100% solvency with zero false-positive circuit breaker activations; the chosen parameters (h_max = 5%,τ_max = 30minutes,θ_price = 50%) were labeled Pareto-optimal. Lowerθ_pricevalues reportedly reduce volume to 80–90%. -
Baseline fragility is documented, not modeled. The paper's framing draws on real incidents: Ronin ($624 million, 5 of 9 validator keys, extracted in under 10 minutes), Wormhole ($326 million), Nomad ($190 million over roughly 2 hours), Orbit Chain ($81 million), and over $1.5B in bridged assets becoming illiquid when Multichain suspended operations in 2023, against cumulative bridge losses of over $2.8 billion since 2021.
-
Internal inconsistencies worth flagging. The main-text statement of Theorem 1 bounds bad debt by
h_max · C_total, while the appendix derivation yields(h_max + 0.005) · C_total= 0.055 · C_total. The circuit-breaker latency threshold is described as 60 minutes in the Orbit example, 24 hours in the circuit-breaker list, and 2 · τ_max = 120 minutes in the liveness proof. The appendices are also truncated in the supplied content, so the manipulation-resistance proof (referenced as Appendix C) could not be verified here.
Methodology in Plain English
The authors start from an analogy to traditional market safeguards — the NYSE's circuit breakers at 7%, 13%, and 20% daily index declines, CME price limits, and the Knight Capital incident where such limits contained losses to $440M over 45 minutes — and ask why bridges, which hold comparable value, have no equivalent graduated defenses.
Their mechanism is to make the bridge's AMM charge for uncertainty. Two signals drive everything: the observed message latency τ (destination receipt time minus source finality time) and price coherence, computed as the deviation between a median-aggregated oracle price (pooling Chainlink, Uniswap V3 TWAP, and Pyth, with a 60-second staleness check) and the last accepted price. Latency is mapped linearly to a haircut between 0.3% at zero delay and 5% at a maximum threshold. The AMM's pricing formula then applies this haircut to the input asset, effectively treating part of the deposit as potential bad debt until time passes.
A Collateral Health Index (locked liquidity times oracle price times one-minus-haircut, divided by outstanding synthetic debt) classifies the system into states: at or above 1.15 it runs normally; between 1.05 and 1.15 it enters Restricted mode with elevated haircuts and doubled slippage; below 1.05 it blocks new mints and pauses outflows. An independent circuit breaker halts everything on a price deviation at or above the configured threshold, on health dropping below 1.05, or on a latency timeout.
For evidence, the authors fork Ethereum mainnet at block 24,102,495 using Foundry, feed in 547 days of one-minute CoinGecko prices (January 2024 – July 2025) plus sanitized Anyswap bridge logs scaled to a 10 BTC / 420,000 USDC pool, and replay 54,700 swaps. They then run 100,000 Monte Carlo simulations with a fixed seed of 42, drawing drawdowns uniformly from 0 to 0.5, latencies log-normally with μ = 5.7 and σ = 2.0 seconds, and swap sizes log-normally with μ = 0 and σ = 1.5 BTC. The full simulation reportedly ran in 0.24 seconds. Contract correctness is checked with a Foundry suite that differentially fuzzes against a standard Uniswap V2 implementation, confirming exact agreement when τ = 0.
Why This Matters
Impact on research. The paper reframes bridge security as an economics problem layered on top of a cryptography problem: it explicitly assumes proofs may be valid yet still economically dangerous during finality windows, and prices that residual risk rather than trying to eliminate it. This is a different research direction from zero-knowledge bridges and light-client proofs, which the authors position as complementary rather than competing. The "safe-to-fail" framing, borrowed from resilience engineering, also gives the DeFi literature a vocabulary for intermediate operational states that most protocols currently lack.
Real-world applications:
- Bridge operators could adopt the latency-indexed haircut as a drop-in economic layer without replacing their existing proof verification.
- Liquidity providers on cross-chain AMMs get a mechanism that keeps capital deployed during volatility rather than withdrawing it — the paper's "flight to safety" result suggests this is the commercially attractive part.
- Lending and synthetic-asset protocols that depend on bridged collateral could use the health-index thresholds as an early-warning signal to reduce exposure before a bridge failure propagates.
- Risk dashboards and monitoring could expose
τand price deviation as public, on-chain indicators of bridge health.
Industry relevance. The counter-intuitive finding that volume rose under protection (104.5% of baseline during stress) is the paper's strongest industry argument: safety features need not be a commercial penalty. That claim matters for bridge operators who fear that pausing or throttling outflows will cede users to competitors.
Future Directions
-
Mechanized formal verification. The authors propose model checking the state machine in TLA+ or Dafny to replace the current hand-written proofs with machine-checked ones — a natural response to the version of Theorem 1 in the appendix diverging from the version in the body.
-
Heterogeneous finality models. The current design parameterizes around one finality assumption (
T_mingiven as roughly 15 minutes for probabilistic chains). Extending to a unified framework covering both probabilistic (Nakamoto) and deterministic (BFT) chains is listed as open work. -
MEV and sandwich resistance. The sandwich attack in the evaluation did not trigger the circuit breaker; it reverted only on slippage. The authors propose integrating private mempools and encrypted transactions to protect large bridge settlements.
-
Resolving the parameter inconsistencies. Reconciling the 60-minute, 120-minute, and 24-hour latency thresholds, and tightening the gap between the stated bad-debt bound (
h_max) and the proven one (h_max + 0.005), would be a straightforward but necessary next step before deployment.
Target Audience
Protocol researchers and engineers working on cross-chain infrastructure, AMM design, or DeFi risk management; security auditors evaluating bridge designs under adversarial relayer assumptions; and graduate students in distributed systems or financial cryptography looking for a concrete example of applying resilience-engineering ideas to permissionless systems. Readers without a background in AMM invariants, oracle mechanics, and Byzantine fault models will find the formal sections difficult, though the motivating incident analysis and the empirical results are accessible on their own.
Authors’ abstract
Cross-chain bridges constitute the single largest vector of systemic risk in Decentralized Finance (DeFi), accounting for over \$2.8 billion in losses since 2021. The fundamental vulnerability lies in the binary nature of existing bridge security models: a bridge is either fully operational or catastrophically compromised, with no intermediate state to contain partial failures. We present ASAS-BridgeAMM, a bridge-coupled automated market maker that introduces Contained…
The full abstract on arXiv