Launch of Cyber Safety Measures for GPT-5.3-Codex Model

On July 9, 2026, ChatGPT Learn announced the introduction of enhanced cybersecurity measures for its GPT-5.3-Codex model. These measures include training the model to refuse malicious requests and rerouting high-risk traffic to a less capable model, GPT-5.2. The initiative aims to improve model performance on cybersecurity tasks while preventing misuse. Developers and security professionals may need to join the Trusted Access for Cyber program to retain advanced capabilities. The company is refining its policies and classifiers to minimize disruptions. The measures are part of a broader effort to balance the dual-use nature of cybersecurity capabilities, ensuring they are available for legitimate research while mitigating risks. The impact on users is expected to be minimal, with ongoing adjustments to the system.

Key Points: • ChatGPT Learn has implemented new cybersecurity safeguards for the GPT-5.3-Codex model. • High-risk requests are rerouted to the less capable GPT-5.2 model to prevent misuse. • Developers can join the Trusted Access for Cyber program to retain advanced capabilities.