Claude Opens Vetted Biology Access With 30-Day Retention

On September 17, 2026, Anthropic opened applications for its Life Sciences Verification Program, a beta for vetted teams and institutions that offers more permissive Claude access for biology work, including project-specific High-risk Use grants that remove life-sciences blocking safeguards. The same launch terms require data associated with program traffic to be retained for 30 days to support offline monitoring.

The program does not give researchers unrestricted Claude access. Applicants must pass institutional checks, grants remain tied to declared use cases, non-biology safeguards stay active, and the initial beta excludes individual plans, third-party platforms and organizations operating under a business associate agreement, or BAA.

Standard and High-risk grants solve different access problems

Every applicant is reviewed for research credentials, security standards and ethical research oversight. After verification, the practical choice is between broad access for routine team workloads and a narrower exception for work that the first tier still blocks.

Standard Use: Intended for most life-science work and capable of covering an approved team’s varied daily workloads. It renews annually and initially covers Mythos 5.1, Opus 5 and Sonnet 5. Its biology classifiers are more permissive than those on generally available models, but biology limits still remain.
High-risk Use: An add-on for a single approved research project that cannot proceed under Standard Use. It receives additional scrutiny, renews every six months and removes all safeguards that block life-sciences requests within the approved scope. At launch, this tier covers Opus 5 and Sonnet 5; High-risk Mythos access remains limited to a small set of additionally vetted entities.

The distinction is one of authorization scope, not researcher seniority. Standard Use can function as a team-level permission for continuing work, whereas a High-risk grant follows a particular project and does not elevate every activity performed by the researcher or institution.

Biology blocks can disappear while other controls remain

High-risk approval removes the safeguards that block life-sciences requests, but cyber classifiers and unrelated protections remain active. Access is also bounded by the use cases described in the application, which should be stated at a high level without sensitive information or intellectual property.

Program traffic is monitored for patterns outside that declared scope. Potentially unauthorized activity can be referred to an organization’s administrators, placing incident triage and remediation partly with the participating institution rather than treating verification as a permanent exemption.

Frontier Warnings’ independent record characterizes the design as a deliberate move from blocking individual biology requests in real time to allowing vetted work and reviewing patterns retrospectively. It also stresses that the launch does not establish that misuse has occurred through the program or that its vetting has failed.

Retrospective monitoring creates the privacy tradeoff

The retention requirement exists because concerning activity can be distributed across multiple requests and sessions rather than appearing in a single prompt. Reviewing those patterns after the fact may reduce interruptions to legitimate research, but it gives Anthropic’s monitoring systems temporary visibility into program activity.

Retained program data is described as compartmentalized, unavailable for model training and inaccessible to members of Anthropic’s life-sciences research teams. Those boundaries limit secondary use, but the public launch material does not provide a field-by-field inventory of the retained data; “logs” therefore refers here to retained traffic data, not to a specified set of log fields.

This arrangement may be incompatible with an institution’s zero-data-retention policy or with workloads containing material that cannot be held under the program’s monitoring terms. Verification expands model access, but it does not remove the need to determine whether a research organization’s contractual and data-handling rules permit that retention.

BAA-enabled organizations and personal accounts are excluded

The initial program is available through Anthropic’s first-party API console and Claude Enterprise or Team plans. Individual Pro and Max subscriptions and third-party platforms are not supported at launch, although expansion to individual access is planned without a published date.

BAA-enabled organizations are also ineligible during the beta. Customers working with protected health information are directed to keep LSVP activity in a separate non-BAA, non-HIPAA organization, rather than moving PHI into the verified environment.

IntuitionLabs’ eligibility analysis underscores that an LSVP grant is permission to use specified models within an approved scope, not evidence of HIPAA, FDA or GxP compliance. For healthcare and pharmaceutical organizations, the unsupported use case is therefore not life-science research itself but running the beta inside the BAA-enabled environment used for protected health information.

Grant selection also varies by product surface. The API and Claude Science support native switching between approved grants, while Claude.ai and Claude Code initially use a preselected default grant, except when Claude Code operates through API authentication.

Eligibility depends on organization, project and data boundary

The applicant must initially be a team or institution rather than an individual subscriber.
The organization must satisfy the review of scientific credentials, security practices and ethical oversight.
Routine work should fit Standard Use; work still blocked there requires a separately reviewed High-risk project grant.
The intended workload must tolerate monitored use and the program’s retention window without placing PHI in the non-HIPAA environment.

Applications are open, but approval, processing time and later expansion are not guaranteed by the public terms. The next unresolved points are when personal and third-party access will arrive, whether the beta can eventually support BAA-enabled organizations, and how Anthropic’s monitoring and grant-portability controls will change as the program develops.