GPT-Rosalind-6 System Card Addendum (deploymentsafety.openai.com)

The recent launch of GPT-Rosalind-6 has introduced a sophisticated reasoning model aimed at advancing biological research, drug discovery, and translational medicine. Available exclusively to qualified scientists and institutions, GPT-Rosalind-6 maintains a low refusal rate for dual-use biological queries while implementing stringent access controls to ensure secure and responsible use. This initiative reflects a careful balance between promoting legitimate scientific inquiry and safeguarding against potential misuse in weaponization. Significantly, GPT-Rosalind-6 employs the same foundational weights as the well-established GPT-6 Astra model, suggesting robust performance and reliability in managing complex biological information. The deployment of this model is classified as high capability within the Biological and Chemical domain but remains cautious to ensure that it does not cross critical thresholds. Under the Preparedness Framework, the system is flagged as critical concerning cybersecurity, while also being evaluated for potential self-improvement implications. This new addition to the GPT family not only enhances the research landscape but also underscores the commitment to ethical standards in AI applications within sensitive fields.