Fewer refusals on dual-use biology, but GPT-Rosalind-6 ships only to vetted labs
Synopsis
GPT-Rosalind-6 is a frontier reasoning model for biology, drug discovery and translational medicine, offered as GPT-Rosalind-6 Research only to qualified scientists, research institutes, companies and government partners. It uses the same weights as the flagship reasoning model GPT-6 Astra, refuses fewer dual-use biological queries while still refusing biological weaponization and retaining refusals on other safety topics. Under the Preparedness Framework the launch is rated High capability in the Biological and Chemical domain, below the Critical threshold, Critical in Cybersecurity and below High in AI Self-Improvement.
Interpretation
A reasoning model aimed at biological research is being released through controlled access rather than to everyone. The releasing organization reviews applicants' proposed uses, security controls and governance practices before granting access, and states that other deployment safeguards remain in place. The releasing organization states that the model has fewer refusals on dual-use biological queries, so it is made available only via trusted access.
The model's capability is not built from scratch but inherited from the flagship reasoning model. GPT-Rosalind-6 uses the same weights as GPT-6 Astra; the difference lies in its positioning toward legitimate biological research, including some dual-use topics, and in where its refusals fall. Under the Preparedness Framework the organization relies on GPT-6 Astra's capability assessments for other Preparedness risks and treats GPT-Rosalind-6 as Critical in Cybersecurity and below High in AI Self-Improvement.
Biological and chemical capability is explicitly placed at High, below the Critical threshold. That rating sits alongside a Critical judgment in cybersecurity, so the same model is positioned differently across risk domains. The rating is given under the organization's Preparedness Framework, which also states that the biological capability assessment, additional safety and alignment evaluations, and deployment safeguards make up this addendum.
Perspective
This addendum covers GPT-Rosalind-6's biological capability assessment, additional safety and alignment evaluations, and deployment safeguards, and includes results for GPT-Rosalind-5.5 where available; it should be read alongside the GPT-6 Astra system card, which describes the parent model's broader capability and safety evaluations. Its intended readers are researchers and engineers concerned with how frontier biological reasoning models are released under control.
Biological and chemical capability is rated High rather than Critical, cybersecurity is rated Critical, and AI Self-Improvement below High; these are the organization's judgments under its own framework. The addendum includes GPT-Rosalind-5.5 comparisons only where available, so not every evaluation has a historical counterpart.