OpenAI Previews GPT-5.6 Sol With Restricted Access and Stronger Cyber Safeguards

OpenAI on Friday released three versions of GPT-5.6, called Sol, Terra, and Luna, as a limited preview to a small number of companies as part of an ongoing engagement with the U.S. government.

While Sol is the latest flagship model and the most powerful, Terra strikes a balance between efficiency and power, and Luna is fine-tuned for speed and affordability.

“GPT‑5.6 Sol launches with our most robust safety stack to date. We strengthened protections for higher-risk activity, sensitive cyber requests, and repeated misuse, and spent multiple weeks finding weaknesses, pressure-testing our system, and hardening it against real-world attacks,” OpenAI said.

The model has also been touted as the “most capable model yet” for cybersecurity, making it much more suitable for vulnerability research and exploitation. On ExploitBench , GPT‑5.6 Sol is competitive with Anthropic Mythos Preview using only about one-third of the output tokens, OpenAI noted.

The goal, it added, is to enable access to legitimate work such as code review, vulnerability research, patch development, debugging, security education, and defensive testing, while enforcing strong guardrails that block offensive activity and swiftly remediating newly discovered jailbreaks. This includes adversarial attempts to jailbreak the model and refuse what it describes as “prohibited cyber assistance.” 

“As these capabilities continue to advance, our priority is to make sure they reach and benefit defenders, who can use these tools to find weaknesses, develop patches, and strengthen systems more broadly,” the artificial intelligence (AI) company explained.

That said, OpenAI is also warning that there may be scenarios during the preview phase where users may encounter safeguards that block or refuse legitimate requests, or have their requests paused for additional review, owing to the “dual-use” nature of the technology.

According to OpenAI’s GPT-5.6 Preview System Card, although the model is more adept at finding vulnerabilities in code and developing exploits, the capabilities do not extend to carrying out autonomous, end-to-end attacks against hardened targets or weaponizing those cyber vulnerabilities in real attacks.

“Separate evaluations examined misaligned behavior in agentic coding tasks and found GPT-5.6 shows a greater tendency than GPT-5.5 to go beyond the user’s intent, including by taking or attempting actions that the user had not asked for, though absolute rates remain low,” it pointed out.

An evaluation of GPT-5.6 Sol against widely deployed hardened software projects using VulnLMP, which is OpenAI’s internal framework designed to test end-to-end exploit chain development against real-world targets, has found the model to produce credible memory safety leads, some of which could lead to disclosure, mutation, or control flow corruption.

“This suggests that substantial parts of real world vulnerability research are becoming increasingly automatable when models are paired with tool use, build systems, and verification infrastructure,” the tech upstart said.

OpenAI intends to make GPT‑5.6 Sol, Terra, and Luna generally available in the coming weeks, and it previewed the model capabilities to the U.S. government. It’s also launching a limited preview for a small group of trusted partners whose participation has been approved by the government before a broader launch.