- The Life Sciences Verification Program allows vetted teams more permissive Claude biology safeguards while requiring Anthropic to retain data associated with flagged activity for 30 days.
- Retained flagged data is compartmentalized, not used for model training, and unavailable to Anthropic’s life‑sciences research teams.
- Verification checks include research credentials, security standards and ethical oversight; applying does not guarantee acceptance.
- At launch, enrollment is confined to Anthropic’s first‑party API console and Claude for Enterprise and Team plans; Pro and Max individual subscriptions are excluded.
- The program can be used through Claude Science, Claude.ai, Claude Code and the API, but approval and grant portability differ across interfaces and products.
- High‑risk Mythos access remains tightly limited; Anthropic is working with the US government to expand access to select partners.
- The beta moves part of safety decisions from per‑request blocks to pattern detection across requests and sessions, allowing approved work to proceed while monitoring for scope violations.
- Unauthorized activity is flagged to organisation administrators, who must triage and remediate within pre‑agreed timeframes.
- The program is designed around three threat paths: compromised access, malicious/coerced insiders, and unintended agent actions.
Useful?