OpenAI

OpenAI Announces GPT-5.4 Thinking with Advanced Cybersecurity Safety Mitigations


Executive Summary

OpenAI has released a System Card for GPT-5.4 Thinking, its latest reasoning model in the GPT-5 series. This release focuses on the model's safety features, highlighting it as the first general-purpose model to implement specific mitigations for high-level cybersecurity capabilities. The approach to cyber safety builds upon the safety work from previous models, such as GPT-5.3 Codex, establishing a new baseline for responsible AI development in advanced systems.

Key Takeaways

* Product Name: GPT-5.4 Thinking (also `gpt-5.4-thinking`).

* Primary Function: The latest advanced reasoning model in the GPT-5 family.

* Key Safety Feature: It is the first general-purpose model with built-in safety mitigations specifically for its "High capability in Cybersecurity."

* Technical Lineage: The cybersecurity safety measures are an evolution of the approaches implemented for GPT-5.3 Codex.

* Model Baseline: GPT-5.4 Thinking is positioned as the successor to GPT-5.2 Thinking.

Strategic Importance

This announcement underscores OpenAI's proactive strategy of addressing the safety risks of increasingly capable models, particularly in sensitive areas like cybersecurity, to build trust and preempt potential misuse.

Original article