OpenAI has introduced three new AI models, GPT 5.6 Sol, GPT 5.6 Terra, and GPT 5.6 Luna, in a limited preview available to a small number of companies as part of an ongoing engagement with the U.S. government. According to the company, Sol serves as the flagship model with the highest overall capability, Terra is designed to balance performance and efficiency, and Luna is optimized for speed and affordability. The preview is currently restricted to a small group of trusted partners whose participation has been approved by the government, while broader availability is expected in the coming weeks.
OpenAI said GPT 5.6 Sol launches with its most extensive safety framework to date, including strengthened protections for higher risk activity, sensitive cyber related requests, and repeated misuse attempts. The company stated that it spent several weeks identifying weaknesses, pressure testing the system, and hardening it against real world attack scenarios before making the preview available. At the same time, OpenAI described Sol as its most capable cybersecurity focused model so far, noting that it performs competitively on ExploitBench against Anthropic Mythos Preview while using roughly one third of the output tokens. The company said the objective is to support legitimate activities such as code review, vulnerability research, patch development, debugging, security education, and defensive testing, while maintaining guardrails that block offensive cyber assistance and rapidly address newly discovered jailbreak techniques.
OpenAI also acknowledged that the dual use nature of advanced cybersecurity AI may result in some legitimate requests being blocked, refused, or paused for additional review during the preview phase. According to the GPT 5.6 Preview System Card, the model has become more effective at identifying software vulnerabilities and developing exploit concepts, but it does not demonstrate autonomous end to end attack capabilities against hardened targets or the ability to weaponize vulnerabilities in real world attacks. The company added that evaluations of agentic coding behavior showed GPT 5.6 has a somewhat greater tendency than GPT 5.5 to take actions beyond the user’s explicit intent, although the overall occurrence rate remains low. Internal testing using VulnLMP, OpenAI’s framework for evaluating exploit chain development against real software targets, found that the model could generate credible memory safety leads that might contribute to vulnerability disclosure, mutation, or control flow corruption research.
The release comes amid increased U.S. government attention on frontier AI systems with advanced cyber capabilities. Earlier this month, U.S. President Donald Trump signed an executive order directing the development of a framework for evaluating AI models and identifying those that qualify as covered frontier models. OpenAI has also recently expanded its cybersecurity initiatives through an updated GPT 5.5 Cyber model for trusted defenders and the Patch the Planet project with Trail of Bits to help secure open source software. The announcement follows a similar move by Anthropic, which received permission to provide its Mythos AI model to approximately 100 trusted companies and federal agencies involved in operating and defending critical infrastructure after earlier restrictions on the cybersecurity focused models were eased.
Follow the SPIN IDG WhatsApp Channel for updates across the Smart Pakistan Insights Network covering all of Pakistan’s technology ecosystem.
