This paper surveys the expanding frontier of smart-contract security beyond simple vulnerability detection, organizing work into four directions: using foundation models for semantic reasoning about vulnerabilities, automated repair with formal guarantees, adversarial learning for robust malicious-contract detection, and real-time exploit detection at blockchain scale. The authors connect these to two recent studies characterizing where analyzers fall short and how to detect malicious transactions in real time, framing the whole as a research roadmap for students designing capstone projects.
Main takeaways:
- Smart-contract security now spans semantic reasoning, automated repair, adversarial robustness, and real-time exploit detection — not just static vulnerability scanners.
- Foundation models are being applied to reason about contract semantics and vulnerabilities in natural language.
- Automated repair aims for formal correctness guarantees, not just heuristic patching.
- Real-time transaction-level detection is now feasible at blockchain scale, moving beyond post-hoc analysis.