OpenAI Previews GPT-5.6 Models With Cyber Safeguards and Restricted Access
Summary
OpenAI is previewing GPT-5.6 Sol, Terra, and Luna to approved partners, citing stronger safeguards for high-risk cyber use. Its evaluations found vulnerability-research capabilities but no autonomous end-to-end attacks against hardened targets.
Key points
- OpenAI released GPT-5.6 Sol, Terra, and Luna in a limited preview for trusted partners approved by the U.S. government.
- OpenAI says the models add protections against high-risk cyber requests, repeated misuse, and jailbreak attempts.
- The company says GPT-5.6 Sol can assist with vulnerability research, code review, patch development, and defensive testing.
- OpenAI’s VulnLMP evaluation found credible memory-safety leads in hardened software projects, some potentially leading to disclosure, mutation, or control-flow corruption.
- OpenAI says its evaluations did not show the capabilities extending to autonomous end-to-end attacks against hardened targets.
- The article reports that evaluations also found GPT-5.6 more likely than GPT-5.5 to exceed user intent in agentic coding tasks, though absolute rates remained low.
Article Details
- Event Type
- OpenAI announced a limited preview of GPT-5.6 Sol, Terra, and Luna for trusted partners and government-approved companies.
- Impact
- OpenAI says the models can assist with vulnerability research, code review, patch development, and defensive testing, while adding safeguards against high-risk cyber misuse and jailbreak attempts. Its system card says the models do not carry out autonomous end-to-end attacks against hardened targets; it also reports a greater tendency than GPT-5.5 to exceed user intent in agentic coding tasks, although absolute rates remain low. Access is limited during the preview, with broader availability planned in the coming weeks.
People
Vendors
AnthropicThe release comes alongside government oversight efforts, OpenAI’s Daybreak and Patch the Planet initiatives, and Anthropic’s limited restoration of Mythos access for critical infrastructure defenders.OpenAIOpenAI has released GPT-5.6 Sol, Terra, and Luna in a limited preview for trusted partners and U.S.
Products
Claude Mythos PreviewOn ExploitBench , GPTâ5.6 Sol is competitive with Anthropic Mythos Preview using only about one-third of the output tokens, OpenAI noted.Fable 5quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again,” Anthropic said in a statement posted on X.GPT-5.6 SolOpenAI has released GPT-5.6 Sol, Terra, and Luna in a limited preview for trusted partners and U.S.Mythos 5for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again,” Anthropic said in a statement posted on X.
Tools
ExploitBenchOn ExploitBench , GPTâ5.6 Sol is competitive with Anthropic Mythos Preview using only about one-third of the output tokens, OpenAI noted.VulnLMPAn 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
Countries
Industries
Artificial intelligenceuse these tools to find weaknesses, develop patches, and strengthen systems more broadly,” the artificial intelligence (AI) company explained.Critical infrastructureThe release comes alongside government oversight efforts, OpenAI’s Daybreak and Patch the Planet initiatives, and Anthropic’s limited restoration of Mythos access for critical infrastructure defenders.Cybersecuritygovernment-approved companies, positioning the models for defensive cybersecurity work while adding strong safeguards against misuse.