OpenAI just unveiled GPT-5.4 Thinking, the newest reasoning model in its GPT-5 lineup, and it comes with a big focus on cybersecurity. While the overall safety approach stays familiar to earlier versions, this is the first general-purpose model to implement serious mitigations for high-level cyber capabilities. The cyber safety measures build on what was already deployed in GPT-5.3 Codex across both ChatGPT and the API. Because there’s no GPT-5.3 Thinking model, the main benchmark to compare against remains GPT-5.2 Thinking. In the blog, the model is also referred to as gpt-5.4-thinking. Why it matters: As AI models get more powerful, integrating cybersecurity protections directly into general-purpose reasoning models signals that safety is no longer an afterthought—it’s a core feature from day one.