- OpenAI released three limited-preview versions of GPT-5.6: the flagship Sol, balanced Terra, and fast/affordable Luna.
- The models, previewed for a small group of government-approved partners, are touted as the “most capable model yet” for cybersecurity, excelling at vulnerability research and exploit development.
- The launch is part of ongoing U.S. government engagement, following a recent executive order by President Donald Trump on AI and cybersecurity.
- Anthropic also recently restored limited access to its powerful Mythos AI model for critical infrastructure defenders.
- OpenAI cautions that despite strengthened safeguards, the “dual-use” nature of the technology means legitimate requests may sometimes be blocked.
OpenAI launched a limited preview of its advanced GPT-5.6 AI models on Friday, focusing on cybersecurity capabilities through government-sanctioned channels. The release, part of an ongoing engagement with the U.S. government, features three versions: the powerful flagship Sol, the balanced Terra, and the speed-optimized Luna.
The company stated GPT‑5.6 Sol launches with its “most robust safety stack to date.” OpenAI aims to enable legitimate security work like vulnerability research and patch development while blocking offensive activity. Consequently, the model is designed as the “most capable model yet” for cybersecurity, according to its announcement.
On ExploitBench , GPT‑5.6 Sol is competitive with Anthropic Mythos Preview using far fewer tokens. Meanwhile, the U.S. government recently permitted Anthropic to release its Mythos model to a group of about 100 trusted critical infrastructure organizations.
OpenAI’s internal evaluations found GPT-5.6 Sol can produce credible memory safety leads against hardened software. The company’s GPT-5.6 Preview System Card noted this suggests real-world vulnerability research is becoming increasingly automatable.
However, OpenAI warns users may encounter blocked requests due to the technology’s “dual-use” nature. The release follows President Donald Trump signing an executive order promoting AI innovation and security.
OpenAI intends to make the models generally available in the coming weeks. This staggered approach ensures controlled access, reflecting heightened regulatory scrutiny over advanced AI capabilities in cybersecurity.
✅ Follow BITNEWSBOT on Telegram, Facebook, LinkedIn, X.com, and Google News for instant updates.
