Anthropic Opens Verified Life-Sciences Access to Claude Models
A beta gives approved teams more permissive biology access through two grant tiers, while retaining cyber controls and adding use-case monitoring and data-retention requirements.
Edited by Tyronne Panaino
Anthropic launched the Life Sciences Verification Program on September 17, opening applications for a beta that gives verified teams and institutions more permissive biology-related access to Claude Mythos, Opus and Sonnet models. The change matters to academic laboratories, biotechnology companies and other life-sciences teams because it creates a formal path to work that Anthropic says is blocked in its generally available Fable models.
This is an access-policy change rather than a new model release. Anthropic says it had already onboarded dozens of organisations through early access and is now accepting applications from the wider life-sciences community. The immediate availability is limited to teams and institutions, with individual Pro and Max support described as a later expansion.
Two grant tiers define the access boundary
The official program announcement describes a verification process that reviews research credentials, security standards and ethical oversight before access is granted. Approved applicants can seek Standard Use or High-risk Use, depending on the work they plan to perform.
Standard Use is intended for broad team activity across research, development, manufacturing, clinical development, quality assurance and related workflows. It currently covers Mythos 5.1, Opus 5 and Sonnet 5, uses classifiers that are more permissive for scientific tasks, and is renewed annually.
High-risk Use is narrower. It applies to one research project, must be renewed every six months and removes the life-sciences request blocks that remain under Standard Use. Anthropic says this tier is available for Opus 5 and Sonnet 5. High-risk access to Mythos remains limited to a small set of additionally vetted entities while Anthropic works with the United States government on broader availability.
Monitoring replaces some real-time blocking
Anthropic says its cyber classifiers remain active across the program. For life-sciences traffic, however, the company is shifting part of the safeguard model from rejecting individual requests in real time to offline monitoring across patterns of activity. Each organisation's access is tied to the use cases declared in its application, and Anthropic says activity outside that scope can be flagged to administrators for investigation.
That design adds a data-handling consequence. The announcement says data associated with flagged LSVP activity must be retained for 30 days so the monitoring process can review it. Anthropic says that retained data is compartmentalised, is not used for model training and is not accessible to its life-sciences research teams.
Availability and compliance limits remain
At launch, LSVP is available through Anthropic's first-party API console and Claude Enterprise and Team plans. It is not yet supported on individual plans or third-party model platforms. The company also says the beta is unavailable to organisations using a business associate agreement; customers handling protected health information would need a separate non-BAA organisation under the vendor's stated launch rules.
The evidence boundary is important. This is a company-authored description of its own access and monitoring system. It establishes the program's stated rules, but it does not independently validate applicant screening, monitoring effectiveness, scientific outcomes or the operational security of participating organisations. The most useful next evidence would be published approval criteria, external evaluation of the monitoring system and documented experience from participating institutions.
Status
Confirmed. Anthropic has opened applications for the beta and documented its access tiers, safeguards, monitoring approach and launch limitations. Internal confidence is medium because the operational claims currently come from Anthropic alone.
Sources
Update note: Last reviewed 2026-09-21. We will revise this post if Anthropic publishes external evaluation, broader availability or material changes to the program's access and data rules.
Sources
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.