Security firm BlockSec has conducted a retest of EVMBench, challenging its assessment of AI's automation capabilities in smart contract audits. By expanding to 26 model configurations and incorporating 22 real attack incidents from February 2026 onwards, BlockSec found that AI had a 0% success rate in exploiting real attacks across 110 tests. However, AI's performance in vulnerability detection was similar to the original report, with some models able to identify known pattern vulnerabilities.