Anthropic opens a biology-permissive Claude to vetted labs

Anthropic opened applications to its Life Sciences Verification Program (LSVP) — a beta trusted-access path that gives verified life-science teams Mythos, Opus, and Sonnet with classifiers more permissive for biology than generally available Fable models. Dozens of orgs already sat in early access; the company says it is now opening to the broader life-science community (teams/institutions first; individual Pro/Max later).

What shipped: Two grant types after credential, security, and ethics oversight review. Standard Use (annual renewal, team-wide) covers most R&D, clinical, manufacturing, and related workflows on Mythos 5.1 / Opus 5 / Sonnet 5 (and future models as they launch). High-risk Use is a six-month, project-scoped add-on that removes life-sciences request blocks for dual-use work; Opus 5 and Sonnet 5 high-risk grants are available now. High-risk Mythos stays limited to a small vetted set while Anthropic says it works with the US government. Cyber classifiers stay on under all LSVP grants. Surfaces: API console, Claude for Enterprise and Team, Claude Science, Claude.ai, Claude Code — not third-party platforms yet; not individual plans yet.

Limits: Beta is not for BAA-enabled orgs — PHI stays on separate non-BAA setups. Enforcement shifts from real-time blocking toward offline monitoring of usage against declared use cases, with 30-day retention of flagged LSVP traffic (compartmentalized; Anthropic says not for training or life-sciences research-team access). Org admins get flags within pre-agreed triage windows.

Who it is for: Academic labs, startups, and pharma that need frontier Claude for biology work blocked on Fable GA — and will accept verification, scoped grants, and shared-responsibility monitoring. Not a consumer toggle.

Risk: Biology dual-use is the main concern. Offline monitoring trades fewer interruptions for retained flagged traffic and org-side remediations. Insider / account-takeover / agent-swarm threat models are Anthropic’s stated targets — treat as company design claims. A verification program that loosens bio refusals is a risk product when grants are diverted.

What’s next: Individual Pro/Max access; Mythos high-risk expansion; third-party platform support; any public incident remediations under LSVP monitoring.

— Rowan Keel