Can Claude Audit Smart Contracts? Zero-Shot Vulnerability Detection Across Five SWC Classes
A zero-shot experiment: one prompt, five known-vulnerable contracts from the SmartBugs Curated benchmark, and an unexpected pattern in how an LLM judges severity.
- Model: Claude Sonnet 4.6 · Claude Pro
- Protocol: Zero-shot · fresh context per contract
- Benchmark: SmartBugs Curated (ICSE 2020)
Claude Sonnet 4.6 found a security bug in each of the five contracts I tested — all of them — without providing a hint, example or setting up anything special. In addition, it placed every single finding at the highest level of risk — Critical — whether the bug was truly critical or not. This combination of great detection ability versus overconfidence in rating is the practical use of this test.
WHY SMART CONTRACT SECURITY IS CATEGORICALLY DIFFERENT
A Smart Contract is a program running on top of the Ethereum Blockchain. It can never be fixed once it has been implemented. It is public, permanent, and will run...
Copyright of this story solely belongs to hackernoon.com. To see the full text click HERE