Anthropic Opens a Life Sciences Verification Program for Vetted Biology Work
On September 17, 2026, Anthropic opened applications to the Life Sciences Verification Program, a beta that gives vetted teams more permissive biology access to Mythos, Opus, and Sonnet. Standard Use renews yearly; High-risk Use is project-scoped, six-month, and still tightly limited on Mythos. Not for BAA or HIPAA orgs, and not on third-party platforms.
TLDR
Anthropic, September 17, 2026: the Life Sciences Verification Program (LSVP) is open for applications, in beta, for teams and institutions. It gives verified life-science organizations access to Mythos, Opus, and Sonnet with biology classifiers that are more permissive than the generally available Fable models. Dozens of organizations were already in early access. Anthropic says it expects to enroll hundreds within the first week.
Two grant types. Standard Use covers most biology R&D, clinical, manufacturing, and regulatory work, can extend to a whole team, and renews once a year. It applies to Mythos 5.1, Opus 5, and Sonnet 5 today, and to later models as they ship. High-risk Use is an add-on for a single research project, renewed every six months, and removes the safeguards that block life-sciences requests. High-risk grants for Opus 5 and Sonnet 5 are available now. High-risk Mythos stays limited to a small set of entities with extra vetting while Anthropic works with the US government. Cyber classifiers stay on under both grants.
Verification reviews research credentials, security standards, and ethical oversight. Access is tied to the use cases named in the grant. Anthropic is shifting this traffic from real-time blocking to offline monitoring, and requires 30-day retention of data tied to flagged activity. That data is compartmentalized: not for training, and not for Anthropic’s life-sciences research teams.
Limits: not on individual Pro or Max plans yet; not on third-party platforms; not available for BAA-enabled orgs, so customers with PHI are told to use separate non-BAA, non-HIPAA orgs. In Claude.ai and Claude Code, only a preselected default grant applies at first, except Claude Code with API authentication.
Product-line de-dupe: not the September 1 Enterprise Frontier Safeguards post, and not the September 10 threat-intelligence report. Those are cited as context. This is the access program.
Why this story matters
Frontier biology is no longer a single on/off classifier. Labs are splitting “daily science” from “named high-risk project,” and putting the monitoring burden on the customer’s own stated scope. The open question is whether offline review plus a 30-day retention rule is enough once agents run long jobs inside a verified org.