Anthropic announced a Life Sciences Verification Program on 17 September, giving vetted life science teams access to its Mythos, Opus and Sonnet models with relaxed safeguards on biology work that general access currently refuses.
There are two tiers. Standard use covers drug discovery, research biology, clinical development and manufacturing, is renewed annually and applies to a whole team. A high-risk add-on covers projects that touch blocked content, is scoped to a single named researcher and is renewed every six months; on Mythos it is limited pending United States government approval. Academic labs, startups, pharmaceutical companies and research organisations are eligible. It is in beta for team and enterprise plans, with individual plans to follow.
The safeguards shift rather than disappear. Monitoring compares usage against a declared use case instead of blocking in real time, prompts are retained for 30 days so they can be analysed offline, and cyber classifiers stay on. Anthropic names three threat models it is designing against: compromised access, insiders and misuse by agents.
On uptake, Anthropic says dozens of organisations were onboarded during early access and that it expects hundreds within the first week. Those are its own figures and nobody has audited them. Xaira Therapeutics, Edison Scientific and Manifold Bio supplied endorsements.
Source: Anthropic, “Introducing the Life Sciences Verification Program”.