OpenAI Unveils GPT-5.6 Sol with Enhanced Cybersecurity Features

In a significant development, OpenAI has released a limited preview of its latest flagship model, GPT-5.6 Sol, to a select group of companies as part of its ongoing collaboration with the U.S. government. This new model boasts the most robust safety features to date, with strengthened protections against high-risk activities, sensitive cyber requests, and repeated misuse. The company has spent several weeks identifying weaknesses, testing its system, and fortifying it against real-world attacks. GPT-5.6 Sol is designed to facilitate legitimate work such as code review, vulnerability research, and patch development, while preventing offensive activity and swiftly addressing any newly discovered vulnerabilities. OpenAI aims to enable defenders to utilize these tools to identify weaknesses, develop patches, and strengthen systems. The model has demonstrated its capabilities in finding vulnerabilities in code and developing exploits, although it is not capable of carrying out autonomous attacks or weaponizing cyber vulnerabilities. The company plans to make GPT-5.6 Sol, along with its Terra and Luna variants, widely available in the coming weeks. This release follows the U.S. government's recent executive order on AI and cybersecurity, which calls for the creation of a framework to evaluate AI models' capabilities. OpenAI's move is part of a broader effort to balance the advancement of AI capabilities with the need for robust cybersecurity measures.