OpenAI has unveiled GPT-5.2, an advanced AI model focused on agentic coding, particularly cybersecurity aspects within large-scale software engineering projects. The new GPT-5.2-Codex model, an extension optimized for agentic building, is designed to support developers and defenders in handling complex, long-term software engineering tasks while reinforcing cybersecurity measures.
The GPT-5.2-Codex model has shown promising results in cybersecurity evaluations, scoring significantly in benchmarks like Capture-the-Flag (CTF) assessments, CVE-Bench, and Cyber Range tests. Notably, the model outperformed previous iterations in these assessments, showcasing its potential to strengthen cybersecurity efforts in software development.
OpenAI is also piloting a program to provide select users access to advanced models for defensive cybersecurity work. By balancing accessibility with safety, the company seeks to empower trusted professionals and organizations to leverage cutting-edge AI technologies for enhancing cyberdefense strategies.
Furthermore, GPT-5.2-Codex has demonstrated improved performance in long-form coding tasks, offering enhanced accuracy and efficiency in handling extensive code changes and large-scale software refactors. The model’s capabilities in supporting real-world software engineering, especially in cybersecurity contexts, mark a significant advancement in AI-driven coding solutions.
Source: VentureBeat
Leave a Reply