Anthropic Opens Claude Mythos to Vetted Biology Teams for Drug Discovery
Anthropic opens vetted access to its restricted Mythos model for biology work, swapping real-time blocking for offline monitoring under a shared-responsibility framework.
- Anthropic launches LSVP beta, opening vetted biology access to Mythos, Opus, and Sonnet.
- Two grant tiers: Standard Use for teams, High-risk Use scoped to single projects, six-month renewal.
- High-risk Mythos access still gated pending US government coordination.
- Safeguards shift from real-time blocking to offline monitoring with 30-day data retention.
- Early adopters include Xaira Therapeutics, Edison Scientific, and Manifold Bio.
- Available via API, Enterprise, and Team plans; no individual Pro/Max or BAA orgs yet.
Anthropic opens Claude Mythos to vetted biology teams
Anthropic has launched the Life Sciences Verification Program (LSVP), a beta that gives approved life sciences organizations access to Claude Mythos, Opus, and Sonnet. The program aims to reduce prompt-level blocks on legitimate drug discovery, pathogen research, and related work while monitoring misuse across sessions.
Claude Mythos had previously remained restricted. When Anthropic released Claude Fable 5 for general use in June, it paired the model with Mythos 5, a controlled version of the same underlying system with some safeguards relaxed. Automated classifiers categorize requests by risk area. Requests involving cybersecurity, biology and chemistry, or model distillation could be routed to the less capable Claude Opus without an explicit model change by the user. LSVP gives verified organizations a formal route around those workflow disruptions.
Access follows the research scope
Organizations apply for grants tied to use cases declared during verification. The program offers two access levels:
| Grant | Scope | Safeguards | Renewal | Models |
|---|---|---|---|---|
| Standard Use | An entire team and its routine workloads | Refined classifiers that allow more scientific requests than generally available models | Annual | Mythos 5.1, Opus 5, and Sonnet 5 |
| High-risk Use | One declared research project | Removes safeguards that block life sciences requests; cybersecurity classifiers remain active | Every six months | Opus 5 and Sonnet 5, with limited Mythos availability |
A lab could use a Standard Use grant for daily research and add a High-risk Use grant for a narrowly defined project, such as studying how human immune pathways recognize a specific family of viral vectors.
High-risk access to Opus 5 and Sonnet 5 is available at launch. Anthropic is working with the U.S. government to expand high-risk Mythos access, which currently remains limited to a small set of organizations that undergo additional vetting.
Monitoring moves beyond one prompt
LSVP centers enforcement on offline monitoring across multiple sessions. Anthropic can examine behavioral patterns associated with insider misuse, account takeover, or groups of AI agents operating outside their approved tasks. General-access safeguards make decisions on individual requests; the beta also evaluates activity against the use case each organization declared when applying.
LSVP traffic is retained for 30 days to support that review. Anthropic says the retained data is compartmentalized, excluded from model training, and inaccessible to its life sciences research teams.
Traffic outside an organization’s approved scope is flagged for its administrators, who must investigate within timeframes agreed with Anthropic. That process shifts part of the enforcement burden to participating organizations and makes account controls, audit procedures, and incident response relevant deployment requirements.
Anthropic’s case for scaling Mythos
Anthropic describes Mythos 5.1 as its strongest model for cybersecurity defense and life sciences research, including threat intelligence, vulnerability discovery, drug discovery, and biodefense screening. Those capabilities are dual-use: the same techniques that support defensive or therapeutic work can also enable harmful activity. Anthropic has therefore limited access through trusted-access programs.
According to Anthropic, a Mythos 5-based drug-design workflow accelerated some tasks by roughly tenfold. The company also says the system matched or exceeded skilled human operators when choosing binding sites, selecting and running protein-design tools, and recovering from tool failures.
Initial participants include Xaira Therapeutics, Edison Scientific, and Manifold Bio. Anthropic expects to enroll hundreds of organizations during the program’s first week and extend access to most of the life sciences community over the following weeks.
Deployment boundaries
- Eligibility: Organizations must undergo verification and declare their intended research uses. Individual Pro and Max accounts are excluded.
- Access points: The beta is available through the API console, Claude for Enterprise, and Team plans.
- Cloud platforms: AWS Bedrock and Google Cloud do not support the program at launch.
- Regulated health data: Organizations operating under a business associate agreement cannot use LSVP. Teams handling protected health information must keep LSVP work in a separate, non-BAA organization and exclude that data from the environment.
- Grant switching: The API and Claude Science support native switching between grants. Claude.ai and Claude Code initially use one preselected default grant.
- Cybersecurity controls: Cyber classifiers remain active under both grant types.
LSVP gives research teams a formal way to prevent legitimate biology prompts from being routed to weaker models. Its practical value will depend on classifier behavior, administrative review costs, and the pace at which Anthropic expands high-risk Mythos access.
Comments
