Independent newsroom

Sunday, November 1, 2026

The File

Every claim traces to a primary source. Uncertainty is labeled. Corrections are public.

AI · Sunday, November 1, 2026

OpenAI posts a safety addendum for a biology model

The File desk · Nov. 1, 2026, 6:29 a.m. ET

Confirmed2 primary sources

Status

Confirmed. OpenAI published the GPT-Rosalind-6 System Card Addendum on October 30, 2026. The capability labels in it are the company's own ratings under a framework it calls Preparedness. This desk did not repeat the tests.
  • Confirmed
    The page says GPT-Rosalind-6 uses the same weights as GPT-6 Astra, with no added training data and no further post-training, and that access is limited to applicants the company approves.
  • Confirmed
    OpenAI says it is treating the launch as High capability in biology and chemistry, below the line it calls Critical, and as Critical in cybersecurity based on the parent model's assessments.
OpenAI published an addendum on October 30 for GPT-Rosalind-6, a model it offers to approved researchers for biology, drug discovery, and medicine. A system card is the company's public account of what a model is for, how the company says it tested the model, and what limits it says it put on who can use it. This addendum is that account for this release.1
The page says the model uses the same weights as GPT-6 Astra. OpenAI wrote, "We did not add training data or perform additional post-training for this release." It says it offers the model as GPT-Rosalind-6 Research to qualified scientists, research institutes, companies, and government partners, and only after it reviews the proposed use, the security controls, and the governance. The company calls that trusted access.1
Why it matters (Estimate). Estimate. A researcher who wants this model has to be approved. A reader who does not apply is not getting a new public chatbot from this page. The addendum is where OpenAI says who may use the model and how it rates the risk. It is not an outside safety ruling. The page says the addendum should be read with the parent system card. This desk opened the addendum and not that parent card.
Under a system OpenAI calls its , the company says it is "treating this launch as High capability in the Biological and Chemical domain, below the threshold for Critical." It also says it will "treat GPT-Rosalind-6 as Critical in Cybersecurity," using assessments of the parent model. Those labels are OpenAI's. This desk did not repeat the tests, and it does not print the biological test scores here.1
How we know

Notes

  1. 1.OpenAI Deployment Safety Hub, "GPT-Rosalind-6 System Card Addendum," published October 30, 2026. https://deploymentsafety.openai.com/gpt-rosalind-6 "GPT-Rosalind-6 is a frontier reasoning model for biology, drug discovery, and translational medicine. We offer it as GPT-Rosalind-6 Research to qualified scientists, research institutes, companies and government partners." "We review applicants' proposed uses, security controls and governance practices before granting access." "GPT-Rosalind-6 uses the same weights as GPT-6 Astra." "We did not add training data or perform additional post-training for this release." "Under our , we are treating this launch as High capability in the Biological and Chemical domain, below the threshold for Critical." "We rely on GPT-6 Astra's capability assessments for other Preparedness risks and treat GPT-Rosalind-6 as Critical in Cybersecurity." The page says the addendum should be read alongside the parent system card. The hub listing also shows the addendum dated Oct. 30, 2026: https://deploymentsafety.openai.com/

What is still unknown or disputed

  • How many applicants have been approved. The addendum does not give a count.
  • What the parent GPT-6 Astra system card says. This desk did not open it for this edition.
  • The October 30 posts on OpenAI's news index. Those article pages returned 403. This story uses the addendum only.

Primary sources

Every claim in this story is drawn from the documents below. If a fetch failed, that is recorded on the card.

  1. Source 1

    GPT-Rosalind-6 System Card Addendum

    OpenAI Deployment Safety Hub · October 30, 2026

    Published October 30, 2026. OpenAI says the model uses the same weights as GPT-6 Astra, is offered through trusted access, and is rated High in biology and chemistry and Critical in cybersecurity under the company's framework.

    https://deploymentsafety.openai.com/gpt-rosalind-6

  2. Source 2

    OpenAI Deployment Safety Hub

    OpenAI · October 30, 2026

    The hub lists GPT-Rosalind-6 System Card Addendum with the date Oct. 30, 2026.

    https://deploymentsafety.openai.com/