Summary
Anthropic has launched a beta program giving verified life-science teams access to more permissive safeguards on its Mythos, Opus and Sonnet models. The program uses use-case verification and offline monitoring, with separate access levels for standard and higher-risk research.
Anthropic has launched the beta Life Sciences Verification Program, giving verified life-science organisations access to its Mythos, Opus and Sonnet models with safeguards that are more permissive for biology-related work than those on its generally available Fable models.
Applications are now open to the wider life-science community. The programme initially targets teams and institutions, including academic laboratories, startups and pharmaceutical companies. Anthropic says it has already onboarded dozens of organisations through an early-access programme.
Two access levels for biological work
The programme is intended for work spanning drug discovery, research biology, clinical development, manufacturing, supply-chain operations, quality assurance, regulatory affairs, investing and diligence. Access is available through Claude Science, Claude.ai, Claude Code and the API.
Applicants undergo a review of their research credentials, security standards and ethical research oversight. Approved organisations can then apply for one of two access types.
Standard Use grants are designed for most life-science research and development workflows. They can cover an entire team, support varied daily workloads and are renewed annually. The grants currently apply to Mythos 5.1, Opus 5 and Sonnet 5, as well as future models as they launch. Their classifiers are intended to permit a broader range of science-related requests than Anthropic’s generally available models.
High-risk Use is an additional grant for work that remains blocked under Standard Use. It removes the safeguards that block life-sciences requests, but is limited to a single research project and must be renewed every six months. Anthropic gives the example of studying how a particular family of viral vectors is recognised by human immune pathways.
High-risk grants for Claude Opus 5 and Claude Sonnet 5 are available at launch. High-risk access for Claude Mythos is being developed with the US government and remains restricted to a small group of entities requiring additional vetting. Cybersecurity safeguards and other protections remain active under the programme.
Monitoring replaces some real-time blocking
Anthropic says the programme was designed around three risks: stolen or compromised accounts being used by another actor, malicious or coerced insiders diverting legitimate access, and autonomous agents taking dangerous actions during long-running or distributed tasks.
The programme ties an organisation’s access to the use cases described in its application. Anthropic will monitor traffic for activity or behaviour patterns outside that approved scope. If unauthorised activity is detected, the company says it can alert organisation administrators, who must respond within pre-agreed timeframes for investigation and remediation.
A major design change is the shift from blocking potentially harmful requests in real time to using offline monitoring to identify suspicious patterns across requests and sessions. Anthropic says this approach is intended to reduce interruptions to legitimate research, while requiring the retention of data associated with flagged activity for 30 days.
The retained data is compartmentalised, cannot be used to train models and cannot be accessed by Anthropic’s life-sciences research teams. The company is also exploring integration with its Enterprise Frontier Safeguards systems for qualifying organisations.
Beta availability and operating limits
LSVP access is currently available through Anthropic’s first-party API console and Claude Enterprise and Team plans. Individual plans are not supported at launch, and the programme is not yet available through third-party platforms. In API and Claude Science, users can switch between grants directly. In Claude.ai and Claude Code, a preselected default grant initially applies, except when Claude Code is authenticated through the API.
The beta is not available to BAA-enabled organisations. Customers handling protected health information must instead use separate non-BAA organisations that are not HIPAA-enabled.
Anthropic says it plans to expand access to individual Pro and Max users and enrol hundreds of organisations, but the current programme remains a controlled, organisation-based access system. Its importance lies in testing whether broader AI capability for biological work can be paired with institutional verification, defined research scopes and post-use monitoring rather than relying only on request-by-request blocking.