On September 17, 2026, Anthropic announced the Life Sciences Verification Program (LSVP), a beta initiative that provides verified life‑science teams and institutions with access to its Mythos, Opus and Sonnet models under a set of safeguards that are more permissive for biology‑related work than the company’s generally available Fable models. Anthropic said it has already onboarded dozens of organizations during an early access phase and is now opening applications to the broader life science community.
Who the program is for and how access is granted
LSVP is intended for a wide range of teams—from academic labs and startups to pharmaceutical companies, clinical developers and manufacturers—who need model assistance for drug discovery, research biology, clinical development, manufacturing and related workflows. Applicants undergo a verification process that reviews research credentials, security standards and ethical research oversight.
Verified organizations may apply for two types of grants:
-
Standard Use (renewed annually): Designed to cover most life‑science activities, including basic research, R&D, supply chain and manufacturing, clinical development, quality assurance, regulatory affairs, investing and diligence. Standard Use grants are available to entire teams for routine workloads and currently apply to Mythos 5.1, Opus 5 and Sonnet 5; they will be extended to future models on release.
-
High‑risk Use (renewed every six months): An add‑on for projects that carry a higher potential for misuse and fall outside Standard Use. High‑risk grants apply to a single research project rather than an entire team and require additional vetting. High‑risk grants are available today for Claude Opus 5 and Claude Sonnet 5; Anthropic is working with the U.S. government to broaden access for Claude Mythos but initially will limit Mythos high‑risk grants to a small set of entities subject to extra review.
Other safeguards, such as cyber classifiers, will continue to operate under LSVP grants.
Shared responsibility and threat models
Anthropic framed the program around protecting against several specific threat models that are especially concerning in biology, where it can be hard to distinguish legitimate work (for example, studying a pathogen to develop vaccines) from malicious intent (for example, seeking to increase transmissibility). The three threat models highlighted are:
- Access compromise: malware or account takeover that diverts access to a bad actor.
- Insider threats: rogue or coerced employees taking malicious action or diverting access.
- Agent misuse: automated agents, especially operating in swarms or on long‑horizon tasks, taking unintended dangerous actions.
To address these risks, Anthropic designed LSVP safeguards around a shared‑responsibility model. Because organizations are vetted for life‑science credibility and oversight, they can specify the safe use cases for their teams or projects when applying. Each entity’s access is tied to those stated use cases, and Anthropic continuously monitors LSVP traffic to detect activity or patterns outside the declared scope. If unauthorized activity is found, Anthropic will flag cases to organization administrators, who are expected to triage and remediate incidents within pre‑agreed timeframes. Use‑case descriptions should be high‑level and must not include sensitive information or intellectual property.
Monitoring approach and data retention
Rather than relying on real‑time blocking of individual requests, LSVP shifts enforcement toward offline monitoring to identify misuse that may be spread across many requests and sessions. This approach reduces interruptions for legitimate work but requires retaining data for review. For LSVP traffic, Anthropic requires 30 days of data retention to allow effective offline monitoring.
Anthropic states that retained data will be strictly compartmentalized, will not be used to train models, and will not be accessible to Anthropic’s life‑sciences research teams. The company is also exploring how LSVP might integrate with features from its Enterprise Frontier Safeguards (EFS) for qualifying organizations.
Availability, products and limitations
Organizations can submit applications to join the LSVP. Anthropic expects to enroll hundreds of organizations in the first week and plans to scale further in the coming weeks to support a majority of the life‑science community.
LSVP is currently available via Anthropic’s first‑party console for API usage and within Claude for Enterprise and Team plans. Individual Pro and Max plans are not yet supported but access will be expanded over time. The program is not yet available on third‑party platforms. As a beta, LSVP is not available to BAA‑enabled organizations; customers with PHI must use separate non‑BAA organizations and non‑HIPAA environments.
In API and Claude Science, users can switch between grants natively. In Claude.ai and Claude Code, an initially preselected default grant applies (except when using Claude Code with API authentication). Anthropic says this default should be sufficient for most users, who will only need a Standard Use grant.
Early reactions from researchers and companies
Several companies featured in the announcement—Xaira Therapeutics, Edison Scientific and Manifold Bio—highlighted their interest in applying Anthropic’s frontier models to drug discovery and related engines, and expressed support for pairing trusted access with accountability to realize AI’s promise in biology.
What’s next
Anthropic said it will continue to refine the program, add products and research collaborations, and improve LSVP features over the coming months with the stated goal of enabling frontier intelligence to advance disease cures and human health.
Key facts at a glance:
- Announcement date: September 17, 2026.
- Models covered today: Mythos 5.1, Opus 5, Sonnet 5.
- Standard Use: annual renewal, team‑level, broad life‑science workflows.
- High‑risk Use: project‑level, six‑month renewal, additional vetting required.
- Data retention for monitoring: 30 days.
- Expected enrollments: hundreds of organizations in the first week.
