GoAT-X: A Graph of Auditing Thoughts for Securing Token Transactions in Cross-Chain Contracts
A new AI auditing framework claims 95% coverage of vulnerable cross-chain smart contracts, potentially automating a task that currently costs millions in manual review.
The Thesis
Cross-chain bridges — software that moves cryptocurrency between separate blockchains — have lost over $2.8 billion to exploits because their security logic spans multiple contracts in ways that are hard for automated tools to follow. GoAT-X proposes a framework that guides a large language model (LLM) through a structured reasoning graph, forcing it to trace data flows and check security properties step by step rather than guessing patterns. The authors claim this approach catches 92% of known vulnerability checkpoints and covers 95% of known vulnerable projects on a benchmark of real-world attacks. The catch is that the benchmark is drawn from already-known exploits, so performance on novel attack classes remains unproven. If the claims hold up at scale, this could reduce the cost and turnaround time for smart contract security audits, a market currently dominated by boutique manual review firms.
Catalyst
LLMs capable of reading and reasoning about multi-file codebases only became practically usable in the past two years, enabling semantic analysis that earlier bytecode-scanning tools could never perform. At the same time, the cross-chain bridge attack surface has grown dramatically as DeFi protocols now routinely span five or more separate blockchains, making manual audits both slower and more expensive. The $2.8 billion in documented losses has created regulatory and institutional pressure for more systematic, auditable security tooling.
What's New
Earlier automated tools for smart contract security — such as Slither and Mythril — work at the bytecode or abstract syntax tree level and match known vulnerability patterns, but they cannot reason about the multi-contract, multi-chain message-passing logic that cross-chain bridges rely on. More recent LLM-based approaches can read source code but tend to produce confident-sounding but incorrect reasoning ('hallucinations') when dependencies span many contracts. GoAT-X addresses this by anchoring LLM reasoning to statically extracted data-flow graphs — meaning the model is only allowed to reason about paths the code actually contains — which the authors argue prevents the model from inventing non-existent execution paths.
The Counter
The benchmark used to validate GoAT-X is built entirely from known, already-disclosed exploits — meaning the system is being tested on cases where the vulnerability type is already understood and catalogued. This is a best-case scenario for any pattern-aware system, and it says little about performance on zero-day attack classes. The 92% recall figure sounds strong, but the paper does not clearly report false-positive rates, which matter enormously in practice: an audit tool that flags hundreds of spurious issues per codebase creates noise that auditors will learn to ignore. LLM-based reasoning over multi-contract codebases is also still computationally expensive and sensitive to prompt design, meaning real-world performance may degrade as contract complexity grows beyond the benchmark cases. Finally, the smart contract security space has seen many 'state of the art' automated tools that performed well on published benchmarks but failed to catch major real-world hacks — the history of Ethereum bridge exploits suggests attackers consistently find logic gaps that static and symbolic tools miss.
Longs
- Coinbase (COIN) — operates cross-chain infrastructure and pays for third-party audits that this tooling could partially displace or accelerate
- CFTC/SEC-adjacent compliance ETFs (e.g., BLKC, DAPP) — broad crypto infrastructure exposure as automated auditing lowers systemic risk
- Palantir (PLTR) — builds AI-driven security analysis pipelines for regulated industries, adjacent market
- CrowdStrike (CRWD) — expanding into blockchain security tooling, direct competitive overlap
Shorts
- Manual smart contract audit boutiques (Trail of Bits, Consensys Diligence, Spearbit) — their moat is the scarcity of expert auditors; automated high-recall tooling compresses that scarcity
- Pattern-matching static analysis vendors — their tools are explicitly called out as insufficient for cross-chain semantics, and a working LLM-based alternative erodes their value proposition
- Cross-chain bridge operators who have positioned their proprietary internal auditing as a competitive differentiator — commoditized tooling removes that advantage
Enablers (Picks & Shovels)
- Static analysis frameworks like Slither (GitHub: crytic/slither) that produce the data-flow graphs GoAT-X anchors its reasoning to
- Open-source LLM inference infrastructure (e.g., vLLM, Ollama) that makes running large code-reading models cost-effective
- Public exploit databases like Rekt.news and DeFiHackLabs that provided the ground-truth benchmark used in evaluation
- Solidity and EVM (Ethereum Virtual Machine) toolchains that expose intermediate representations parseable by static analysis
Private Watchlist
- Trail of Bits — leading smart contract audit firm that could adopt or compete with this tooling
- Certik — large-scale automated + manual blockchain security auditor, direct domain overlap
- Chainalysis — blockchain analytics firm with incentive to integrate vulnerability detection
- OpenZeppelin — open-source smart contract security library maintainer, natural distribution channel
Resources
The Paper
Cross-chain bridges, the critical infrastructure of the multi-chain ecosystem, have become a primary target for attackers, resulting in over $2.8 billion in losses due to subtle implementation flaws. Existing defenses, such as bytecode-level static analysis, are ill-equipped to handle the semantic complexity of cross-chain interactions, while LLM-based approaches, which can understand source code, struggle with hallucinatory reasoning over complex, multi-contract dependencies. In this paper, we propose GoAT-X, a framework that shifts automated cross-chain smart contract codebases auditing from heuristic pattern matching toward systematic first-principles verification. GoAT-X structures the audit process as a Graph of Auditing Thoughts, explicitly mirroring how human experts decompose, reason about, and validate security logic. By anchoring LLM reasoning in statically extracted data flows and explicitly linking abstract security properties to concrete code implementations, the framework constrains semantic reasoning within well-defined structural and state boundaries. Within this constrained space, GoAT-X treats missing constraints and adversarial bypass paths in cross-chain logic as first-class vulnerability targets and dynamically explores reasoning paths to identify exploitable semantic gaps. We evaluate GoAT-X on a comprehensive benchmark covering all known cross-chain token transaction attacks. GoAT-X achieves 92% recall on fine-grained audit points and 95% coverage of vulnerable projects, while identifying 117 confirmed risks in the wild with low operational cost, establishing a new standard for scalable, logic-driven cross-chain security.