Automatic synthesis of attack-stopping smart-contract invariants

Develop methods that automatically find correctly placed smart-contract invariants capable of stopping real-world exploits, beyond the limited success achieved by existing invariant-generation tools.

Background

The paper evaluates whether manually authored program invariants can block real Ethereum exploits and whether automated tools can discover those invariants. Using the InvariantEval benchmark of 28 real-world attacks, the authors find that every human-authored invariant blocks its corresponding attack, while FLAMES, InvCon, and InvCon+ together recover only 2 of the 28 attack-stopping invariants. Thus, although invariant-based hardening is empirically effective when the appropriate guard is known, automatically discovering the security-critical invariant remains unresolved.