According to PANews news on June 2, according to Blockworks, executives of blockchain security company OpenZeppelin said that a recent company experiment proved the continued demand for human auditors. OpenZeppelin conducted a study that tested whether GPT-4 could identify various smart contract vulnerabilities in 28 Ethernaut challenges.

Of the 23 Ethernaut challenges introduced before GPT-4's training data cutoff (September 2021), the AI tool successfully solved 19, then failed tasks in the five challenges introduced after September 2021 4. It is reported that OpenAI pointed out on its website that this AI tool "generally lacks" understanding of events after September 2021 and "cannot learn from experience."
"In some cases, GPT-4 was able to correctly identify a vulnerability, but failed to explain the correct attack vector or propose a solution; Vulnerabilities that exist. For problems solved by AI tools, security experts must provide additional hints to guide them to the correct solution. Extensive security knowledge is essential to assess whether the answers provided by AI are accurate; It will lead to a shift in the work of developers and inspire innovative useful tools to improve efficiency, but it is unlikely that AI will replace human auditors in the near future.”