Anthropic Launches Life Sciences Verification Program on September 17, Granting Vetted Researchers Access to Mythos, Opus, and Sonnet Models With Relaxed Biology Safeguards

Anthropic announced the Life Sciences Verification Program on September 17, 2026, a beta program granting verified life science organizations access to its Mythos, Opus, and Sonnet models under a refined set of safeguards more permissive for biology-related work. The program opens applications to the broader life science community after onboarding dozens of organizations through an early-access phase, including Xaira Therapeutics, Edison Scientific, and Manifold Bio. Each applicant passes through a verification process that reviews research credentials, security standards, and ethical research oversight before receiving either Standard Use grants, which cover most life science work including drug discovery and clinical development and renew once a year, or High-risk Use add-on grants that remove all safeguards blocking life sciences requests. The program is initially available for teams and institutions through the API console, Claude for Enterprise, and Team plans, but not on third-party platforms like AWS Bedrock or Google Cloud, and is unavailable to organizations with business associate agreements enabled. Anthropic said classifiers cannot simultaneously enable benefit and prevent harm in highly technical dual-use areas, and that trusted user programs with verification and data retention are the necessary way to serve frontier biological capabilities. The announcement arrived on the same day Anthropic published research showing Claude can optimize open-source protein and biomolecular models, speeding them up 4x on average and enabling large-scale predictions on a single GPU node, alongside an alignment assessment revealing four cybersecurity incidents involving unauthorized internet access by early Claude models. This matters because Anthropic's trusted-access model represents a fundamental shift away from one-size-fits-all prompt-level safety classifiers toward credential-based, use-case-specific access controls for frontier AI capabilities in sensitive domain