OpenAI

OpenAI Launches GPT-5.2-Codex for Agentic Coding and Defensive Cybersecurity


Executive Summary

OpenAI has released GPT-5.2-Codex, a new AI model optimized for complex, agentic software engineering and defensive cybersecurity tasks. This specialized version of GPT-5.2 offers improved performance for long-running coding sessions, large-scale code changes, and significantly enhanced cybersecurity capabilities. The model is available immediately for paid ChatGPT users, with API access planned, alongside a new invite-only trusted access pilot for vetted security professionals to safely utilize its advanced features.

Key Takeaways

* Product Name: GPT-5.2-Codex

* Primary Function: An agentic AI model for advanced, real-world software engineering and defensive cybersecurity.

* Key Capabilities:

* Improved long-context understanding and context compaction for extended coding tasks.

* Stronger performance on large code changes like refactors and migrations.

* Enhanced reliability in Windows environments.

* State-of-the-art performance on SWE-Bench Pro and Terminal-Bench 2.0 benchmarks.

* Advanced vision for interpreting technical diagrams, screenshots, and UI mocks.

* Cybersecurity Focus: The model possesses significantly stronger cybersecurity skills, prompting a cautious deployment strategy. A real-world example is cited where a predecessor model was used to discover new vulnerabilities in React.

* Availability:

* General: Available now in all Codex surfaces for paid ChatGPT users.

* API: Access for API users is planned for the coming weeks.

* Trusted Access Pilot: OpenAI is launching an invite-only program for vetted security professionals and organizations to access the most capable models for defensive cybersecurity work, such as vulnerability research and authorized red-teaming.

Strategic Importance

This release marks a strategic move by OpenAI to create highly specialized models for professional verticals while proactively managing dual-use risks. The tiered access model, particularly the trusted access pilot, establishes a framework for safely deploying powerful AI capabilities to domain experts, setting a potential industry precedent for responsible scaling.

Original article