Video Preview
🎧 Podcast episode for paid subscribers only. Also available on Apple Podcasts and Spotify.
To listen to paid episodes in Apple or Spotify, link your Substack subscription via the show settings on those platforms (instructions inside the Substack app under Subscriptions → Podcast).
Abstract
What actually launched on September 17
Two grant types, one very important line between them
The threat model, aka why the prompt stopped being the unit of trust
From real-time blocking to offline monitoring, and the 30-day catch
Who is locked out for now: BAA orgs, individuals, and third-party clouds
How it stacks up against the other gated doors
What this signals for pharma, biotech, and the people writing checks
The scorecard
On September 17, 2026, Anthropic opened applications for the Life Sciences Verification Program (LSVP), a gated access tier that lets vetted life science organizations use Mythos 5.1, Opus 5, and Sonnet 5 with safeguards that are looser for biology than the generally available Fable models. The program splits access into a team-wide, annually renewed Standard Use grant and a project-scoped, six-month High-risk Use grant that removes all life-science blocking for Opus 5 and Sonnet 5, with Mythos 5.1 High-risk access still limited to a small number of additionally vetted entities. The architectural move underneath is a shift from per-request classifier refusals to offline pattern monitoring tied to each org’s self-declared use cases, which requires 30 days of data retention on LSVP traffic. Cyber classifiers stay on. BAA-enabled orgs are excluded for now, individual Pro and Max plans are not supported, and third-party clouds are out of scope. This essay walks through the mechanics, the threat model, the friction points, how it compares to OpenAI’s trusted access programs and the pre-AI precedents in biosafety governance, and what it means for anyone building, buying, or investing in AI for drug discovery.
