Anthropic launches a life sciences verification program

Anthropic launched a Life Sciences Verification Program, under which vetted biology organisations can get eased safety classifiers on Claude for drug discovery, research biology, clinical development and manufacturing work. The company’s standard classifiers are tuned to block queries that combine biological information in ways that could indicate misuse; Anthropic said those same classifiers also catch legitimate research queries, and that verification lets it relax them for applicants whose credentials, security practices and ethical oversight have been reviewed.

The program has two tiers. Standard-use verification, renewed annually, applies refined classifiers across an organisation’s day-to-day workflows on Claude Science, Claude.ai, Claude Code and the API. High-risk verification, renewed every six months and granted project by project, removes life-sciences safeguards more fully for sensitive research and has limited availability on Claude Mythos. Rather than blocking flagged activity in real time, Anthropic said verified use is monitored through offline review, shifting enforcement from the point of query to after-the-fact oversight backed by the applicant’s own compliance obligations.

The program is a beta, and Anthropic framed it as balancing legitimate scientific work against biosecurity risks such as compromised accounts, insider threats and misuse by autonomous agents. It is a departure from labs’ usual practice of applying one safety policy uniformly across all users: it formalises a verified professional class that gets a different, looser standard, betting that vetting and monitoring can substitute for blanket restriction without materially raising the risk of misuse.