GPT-Rosalind-6 System Card Addendum
Reported by Hacker News1 source ↓
An addendum describes GPT-Rosalind-6’s biological capability assessment, safety evaluations and access safeguards. The model is offered to qualified research users through trusted access.
Key points
- GPT-Rosalind-6 is available to qualified scientists, research institutes, companies and government partners
- Applicants are reviewed for proposed uses, security controls and governance practices
- The addendum rates the model High in biological and chemical capabilities, below Critical
- The model uses the same weights as GPT-6 Astra
Details
The addendum says GPT-Rosalind-6 is designed for biology, drug discovery and translational medicine, including some dual-use research. It says the model refuses requests for biological weaponization and retains refusals on other safety topics, while having fewer refusals on dual-use biological queries.
Access is limited to trusted users. The addendum says applicants’ proposed uses, security controls and governance practices are reviewed before access is granted, and that other deployment safeguards remain in place.
Under the Preparedness Framework, GPT-Rosalind-6 is rated High in the Biological and Chemical domain, below the Critical threshold. The addendum relies on GPT-6 Astra’s assessments for other Preparedness risks, rates Rosalind Critical in Cybersecurity and below High in AI Self-Improvement, and includes GPT-Rosalind-5.5 comparisons where available.
Written from coverage as of