OpenAI's GPT-5.6, developed under the project code names Sol, Terra, and Luna, became available for unrestricted global access on October 15 after the Trump administration lifted export restrictions that had applied to the model since its initial domestic launch. The change expands access to a model that US regulators had restricted specifically because of its demonstrated capability in software vulnerability identification.
A different model from GPT-6
GPT-5.6 is distinct from GPT-6, which OpenAI made available through its API earlier in October 2026. GPT-5.6 was developed as a specialized model with particular strength in code analysis and vulnerability research, properties that prompted the original export restriction. GPT-6 is a broader general-purpose frontier model. Both are now available globally.
Why regulators originally restricted it
Bloomberg Law reporting from the original restriction period cited US export control concerns around software vulnerability tools. OpenAI's regulatory filings described GPT-5.6 as having 'unprecedented ability to identify software vulnerabilities,' a characterization that placed it in a policy category alongside offensive security tools subject to dual-use export rules. The Trump administration's decision to lift the restriction reflected a policy judgment that restricting US models placed American AI companies at a competitive disadvantage without meaningfully reducing global access to equivalent capability.
Security implications for practitioners
For defensive security teams, GPT-5.6's global availability raises two practical questions. First, the same vulnerability-identification capability that prompted the restriction is now available to researchers and developers worldwide, including in jurisdictions with active offensive programs. Second, the model's availability through OpenAI's API creates new inputs for automated vulnerability discovery pipelines, potentially compressing the time between vulnerability existence and the first exploitation attempt. Security teams should revisit their AI acceptable use policies to account for both the defensive utility and the offensive potential of models with explicit vulnerability research capabilities.
Coordinated regulatory lift
Anthropic restored unrestricted global access to Fable 5 and Mythos 5 on the same day, suggesting the regulatory lift was coordinated across leading frontier model providers rather than specific to OpenAI. The simultaneous announcement is consistent with the pattern of the Trump administration managing frontier AI policy through bilateral engagement with major labs rather than formal rulemaking.