Anthropic's Life Sciences Verification Program opens beta access to Claude Mythos, Opus, and Sonnet for biology research
Verified life-science organisations can now apply for relaxed Claude safeguards across claude.ai, the API, and enterprise plans from 17 September 2026.
Anthropic opened applications on 17 September 2026 for its Life Sciences Verification Program (LSVP), a verified-access scheme that gives qualifying biology organisations a more permissive set of Claude safeguards for research work. The programme is in beta, covering teams and institutions rather than individual users, and is available through Claude.ai, the API, and Enterprise and Team plans.
What the programme actually unlocks
Anthropic’s generally available Fable models carry classifiers that block certain biology, chemistry, and cybersecurity-related requests. When those classifiers fire, the request is handed off to a less capable model. The LSVP changes that for verified organisations: their biology-related queries are handled by the full versions of Claude Mythos 5.1, Opus 5, and Sonnet 5 without the Fable-level biology blocks in the way.
Non-biology safeguards remain in place. If your team is using Claude for drug discovery but also occasionally asks about cybersecurity topics, those requests still go through the standard classifiers.
Two tiers of access
Once your organisation is verified, you can apply for one of two grant types:
Standard Use covers the majority of life science workloads, including most biology research and development tasks. Grants are team-wide and renew annually. They apply to Mythos 5.1, Opus 5, and Sonnet 5 from launch, and to future models as they become available.
High-risk Use is project-scoped, lasts six months, and removes all life-science blocking for Opus 5 and Sonnet 5. Mythos 5.1 High-risk access is more tightly controlled: at launch it remains limited to a small set of additionally vetted entities, with Anthropic working alongside the US government on broader availability.
The verification process itself reviews research credentials, security standards, and ethical research oversight before a grant is issued.
What this means for your organisation
If your team does biology research and has run into Claude refusing or watering down responses on legitimate scientific questions, this is the path to a better experience. Early adopters include Xaira Therapeutics, Edison Scientific, and Manifold Bio, and Anthropic reported that a Mythos 5-based drug-design workflow accelerated certain tasks by roughly tenfold, with the model matching or exceeding skilled human operators on binding-site selection, protein-design tool use, and error recovery.
The practical value is clearest for teams doing drug discovery, research biology, clinical development, or manufacturing work where the Fable classifiers have been a recurring obstacle.
For most organisations, the Standard Use grant is the right starting point. High-risk Use is for situations where even the loosened Standard settings are insufficient, and the additional scrutiny involved reflects the sensitivity of the work.
What you need to know before applying
A few constraints are worth being aware of before you start the process:
US-only at launch. Anthropic is initially limiting the programme to US-based organisations and says it is working to expand internationally.
No individual accounts. Pro and Max plan users are excluded from the beta. Anthropic says it will expand to individual plans over time.
Not available on third-party clouds. AWS Bedrock and Google Cloud do not support the LSVP at launch. Access is through Anthropic’s own console, Claude.ai, and Enterprise or Team plans.
BAA-enabled organisations excluded from beta. If your organisation has a Business Associate Agreement with Anthropic, you will need to wait for a later phase.
Data retention applies. LSVP traffic is retained for 30 days to support Anthropic’s offline monitoring. The retained data is compartmentalised, excluded from model training, and kept away from Anthropic’s own life sciences research teams.
The monitoring architecture is worth understanding: rather than blocking individual requests in real time, LSVP shifts to offline pattern monitoring tied to each organisation’s declared use cases. That is what makes broader access workable without removing oversight entirely.
The broader context
Anthropic has been building its life sciences presence since at least October 2025, when it launched Claude for Life Sciences. The LSVP formalises that trajectory. The company has also established partnerships with Novo Nordisk, the Allen Institute, and Howard Hughes Medical Institute, and is co-sponsoring a protein design competition with Adaptyv Bio backed by up to $1 million in Claude credits.
The programme sits alongside Anthropic’s existing AI for Science credits offering, which provides researchers with credits ranging from $20,000 to $50,000. The two programmes are complementary rather than mutually exclusive.
Pricing for Claude Mythos 5.1 starts at $10 per million input tokens and $50 per million output tokens.
Anthropic has also been candid about why the verification layer exists. Its recent misuse report detailed five cases in which researchers used Claude for advanced biological work that could have served harmful as well as beneficial purposes. The LSVP is partly a response to that: a way to extend access to legitimate researchers while maintaining visibility into how that access is being used.
If your organisation qualifies, applications are open now at anthropic.com/news/life-sciences-verification-program.