Catenaa, Monday, April 27, 2026- AI security firm Cecuro has reported a significant performance lead over competing systems in a widely used smart contract vulnerability benchmark developed with OpenAI and other research groups.
The company said its multi-agent AI system identified 87.7 percent of high-severity vulnerabilities in EVMBench, an open-source evaluation framework for blockchain security. The benchmark tests AI models on detecting, exploiting, and patching weaknesses in Ethereum-compatible smart contracts.
EVMBench includes 120 high-severity vulnerabilities across 40 real-world audit cases. Cecuro’s system detected 101 of those issues, nearly doubling the performance of its closest competitor in the evaluation.
By comparison, Anthropic’s Claude Opus 4.6 scored 45.6 percent in the same test. Other models, including GPT-based systems and Gemini 3 Pro, recorded significantly lower detection rates.
The benchmark was developed by OpenAI, Paradigm, and OtterSec and has quickly become a reference point for measuring AI capabilities in blockchain security analysis. ([chainwire.org](https://www.chainwire.org/press-release/ai-audit-firm-cecuro-outperforms-rival-benchmark?utm_source=chatgpt.com))
The results arrive amid rising concerns about automated attacks on decentralized finance systems. Research cited in the report suggests AI-powered tools can now scan smart contracts for vulnerabilities at extremely low cost, increasing pressure on blockchain security systems.
Earlier studies have shown that offensive AI tools can identify weaknesses in thousands of contracts for under a few thousand dollars, while exploit capabilities continue to improve rapidly.
Industry data cited in the report indicates that billions of dollars in crypto assets were lost to smart contract vulnerabilities in recent years, highlighting the scale of the security challenge.
Implications
The performance gap between specialized AI security systems and general-purpose models suggests that domain-specific training remains critical in blockchain security.
General AI models performed significantly worse in structured vulnerability detection tasks, particularly in decentralized finance environments where exploits often involve complex interactions between multiple contracts.
Cecuro said its approach uses multiple AI agents working together, combining structured analysis with domain-specific security logic to improve detection accuracy.
The findings raise concerns that offensive AI tools may be advancing faster than defensive systems. This could increase the risk of automated exploits targeting smart contracts across decentralized finance platforms.
At the same time, the results suggest that specialized AI security tools may become a key layer of infrastructure for blockchain networks, particularly as financial value locked in smart contracts continues to grow.
Industry observers say the next challenge will be ensuring defensive AI systems can keep pace with increasingly automated attack methods.
EVMBench was introduced in early 2026 as a standardized testing environment for evaluating AI performance in blockchain security tasks. It uses real-world exploit data and controlled execution environments to measure accuracy and reliability.
