Anthropic on Thursday announced the Life Sciences Verification Program, a new tiered access system that gives vetted life science professionals access to its Claude models with safeguards that are deliberately more permissive for biology-related work than what general users get.

The program, known as LSVP, opens applications to the broader life science community after an early-access phase in which the company says it has already onboarded dozens of organizations. It is launching in beta for teams and institutions first, with access for individual Pro and Max subscribers planned over time. For more context on this story, see our ongoing AI trends.

What the program actually unlocks

The core problem LSVP addresses is a familiar one for AI labs with biology-adjacent customers: Anthropic's generally available models — the company's Fable line — block many life science requests outright, because the same knowledge that helps a cancer researcher design an experiment can, in principle, help someone engineer something dangerous.

Under LSVP, verified teams get access to Anthropic's Mythos, Opus and Sonnet models with "refined classifiers that are more permissive for science tasks than our generally available models," the company said in its announcement. The intent is to enable work that is currently blocked in general models: drug discovery, research biology, clinical development, and manufacturing workflows, spanning everything from basic science and R&D to supply chain, quality assurance, regulatory affairs, and even investment diligence.

In practice, that means a pharma company's research team could use Claude to reason through assay design or molecule optimization workflows without tripping safety classifiers tuned for the general public — tasks the company explicitly says are blocked for regular users today.

Two tiers of access, two levels of vetting

The program is built around two grant types with materially different rules.

Standard Use grants cover most life science work. They are issued to entire teams, renewed annually, and apply to Mythos 5.1, Opus 5 and Sonnet 5 today, plus future models as they launch. These grants loosen biology-related safeguards while keeping the rest of Anthropic's safety infrastructure intact. According to Anthropic, Standard Use is expected to cover the majority of access needs — day-to-day workloads across academic labs, startups, and large pharmaceutical companies. High-risk Use grants are the more contentious tier. Designed for teams whose work touches areas blocked even under Standard Use — Anthropic offers the example of characterizing how a specific family of viral vectors is recognized by human immune pathways — these grants "remove all safeguards that block life sciences requests." They attach to a single research project rather than a whole team, and must be renewed every six months.

High-risk grants for Opus 5 and Sonnet 5 are available at launch. Grants for Claude Mythos, the company's most capable model, remain restricted to "a small set of entities with additional vetting" while Anthropic works with the US government to expand availability — an acknowledgment that the frontier model sits in a different risk class.

How verification works

Every applicant goes through a verification process that includes review of research credentials, security standards, and ethical research oversight before any grant is issued, according to the announcement. That puts the burden of proof on institutions rather than the model: a verified team's standing, security posture and oversight structures vouch for its requests, rather than relying solely on classifiers guessing at intent from the text of a prompt.

Verified teams can use their grants across Anthropic's product surfaces, including Claude Science, Claude.ai, Claude Code, and the API.

What stays locked down

The loosening is scoped to biology. Anthropic notes that all other safeguards — cyber classifiers in particular — remain in place under LSVP grants. A verified cancer lab does not get a freer hand on offensive security tasks, and a High-risk Use grant for a viral vector study does not transform into general frontier access.

Why it matters

The announcement lands in the middle of an escalating argument about AI's role in biology. Recent studies demonstrating that AI models can design viable virus genomes — including work published in Science earlier this year — have intensified scrutiny of exactly how much biological capability labs should expose, and to whom. Critics argue that any permissiveness is a gamble; researchers counter that over-restrictive classifiers block legitimate medical work and push scientists toward less capable or less transparent tools.

Anthropic's tiered answer is effectively a bet that identity verification plus institutional accountability can substitute for blanket restriction. It is a notable shift from the industry's default posture of filtering everything and asking questions never — and a pragmatic recognition that its biggest science customers were the ones most likely to hit a wall.

The economics point the same direction. Pharmaceuticals, biotech and academic research are among the deepest-pocketed potential customers of frontier AI, and Claude Science already competes for those workloads. A program that lets Merck or a Stanford lab actually run their workflows on Claude — rather than abandonment at the first blocked prompt — is a commercial weapon as much as a safety policy.

Whether the bet holds depends on the vetting. If verification holds and misuse stays theoretical, LSVP will likely become the template other labs copy. If it fails, the program hands regulators a concrete example of what loosened safeguards look like in practice — at exactly the moment policymakers in Washington and London are asking AI's leaders to slow down.

---

Stay Ahead of AI

Get the latest AI news, analysis, and breakthroughs — all in one place.

Read more AI news →