EVOHUNT AI beats OpenAI's Codex Security with 11.3% bug detection rate
A $1,400 experiment shows open-source AI outperforms a commercial security tool.
On June 23, 2026, a research team announced that their EVOHUNT AI system outperformed OpenAI's commercial Codex Security product in a bug-hunting experiment. EVOHUNT detected vulnerabilities in 11.3% of cases compared to Codex Security's 9.2% across 371 test cases, with the entire experiment costing just $1,400. The key innovation behind EVOHUNT is its ability to evolve an external 'playbook' written in plain text while the underlying AI model remains fixed. This approach allows the system to adapt its strategy without retraining the model, making it highly efficient and cost-effective.
The results challenge the assumption that commercial, proprietary AI security tools are inherently superior. By using an open-source methodology and a fixed model, EVOHUNT demonstrates that iterative learning through an external playbook can surpass the performance of a heavily optimized product like Codex Security. The low cost of the experiment ($1,400) further highlights the accessibility of advanced AI security research. This breakthrough could democratize vulnerability detection, enabling smaller teams and organizations to deploy effective bug-hunting systems without relying on expensive commercial tools.
- EVOHUNT found 11.3% of vulnerabilities vs Codex Security's 9.2% across 371 test cases.
- The experiment cost only $1,400, a fraction of comparable commercial deployments.
- EVOHUNT evolves an external playbook in plain text while keeping the AI model fixed.
Why It Matters
Open-source AI security tools can challenge commercial products at a fraction of the cost.