Biological capability is now being generated faster than anyone can verify its safety. AI agents are outperforming PhD-level virologists on technical laboratory questions in the virologists' own subfields. Sequence-based screening and customer verification, the cornerstones of biosecurity since 2009, are no longer sufficient as standalone safeguards against AI-assisted design, especially given unclear liability across the design-synthesis-use pathway.

The stack is surprisingly brittle. Agents can plan and execute multi-step design workflows while concealing their work from trace-based monitoring, poisoned data can do meaningful damage at as little as 1% of inputs, and red-teaming results go stale the moment a new model is released.

We're looking for teams to build the verification and containment layer for this world. The question is not whether AI creates biological risk, but how to deliver guarantees that are auditable, testable, and durable as capability keeps compounding.

While AI is reshaping healthcare delivery, one frontier remains underexplored: the physical work of caring for patients and meaningfully improving their quality of life, especially when it comes to movement.

We're looking for teams to build robots, wearables, and prosthetics, along with the intelligence behind them, that operate in the physical world of hospitals, homes, and rehab settings. Areas of interest include, but aren't limited to, perioperative automation, assistive mobility, neuromodulation, patient-facing robot design, and simulation environments.

Above all, we want embodiment: systems that sense, move, and respond continuously alongside patients and clinicians, rather than sitting in a chat window. Many past solutions were held back by hardware constraints and poor economics, so show us why now is different.

We are interested in how modern agentic workflows can supercharge drug discovery, especially approaches that are mindful of the current limitations of AI in this field. We suspect that the most promising approaches will involve tightening information loops: a new paper in one field triggers drug discovery signals in another; expanded datasets improve agentic drug repurposing; and precision therapeutics allow for drugs tailored to a patient’s entire disease profile, rather than a single pathway or mutation.

We think rare pediatric diseases are a particularly promising use-case, with multiple low-hanging fruit, opportunities for long-term impact, and regulatory incentives like Priority Review Vouchers (PRVs) to reward successful development.

We are also interested in workflows that use agents to enhance the patient journey, such as ambient scribes/agents tailored to an individual’s rare disease that help patients and their families understand a diagnosis or navigate the clinical trial landscape.

As interest grows in longevity innovations, we need AI that can help make sense of it all.  We see a critical need for AI systems that translate how a careful scientist reads a paper into rubrics a machine can run. These rubrics would cover study design, trust tiers, sponsor independence, aggregation of conflicting evidence, and safety checks against federal sources.

Systems like these could help patients and providers turn everyday questions about supplements, diagnostics, and longevity interventions into honest, traceable answers grounded in the scientific literature.

We also see this as a strong starting point for improving healthcare AI evaluations more broadly.

By training on the sea of publicly available health data, LLMs have turned loosely structured knowledge into coordinated intelligence.

The next generation of agents will need more private data: patient records, internal company data (such as hospital system utilization), or data at the edge from phones and logging systems. Getting there will take a combination of skills, including navigating regulations on protected health information (PHI), extracting insights in place from firewalled systems, and bootstrapping statistically grounded synthetic datasets.

Private data is the next frontier, and for teams willing to tackle the thorniest problems, the bigger the mess, the larger the moat. We're interested in new ways to source complementary datasets that expand the alternative data market, as well as tools that help companies unlock more of their own data safely and compliantly.

We’re interested in all company ideas that are exploring AI at the intersection of healthcare or life sciences. You should apply even if you’re not sure your idea is a good fit.