Insilica CEO Co-Authors Framework for Evidence-Based AI in Toxicology

Insilica announced the publication of a new framework called Evidence-based AI, according to a press release. The work, published in Frontiers in Artificial Intelligence, was co-authored by Insilica CEO Dr. Thomas Luechtefeld and Dr. Thomas Hartung of the Johns Hopkins Bloomberg School of Public Health.

The framework applies the rigorous standards of evidence-based medicine and toxicology to AI systems. Its central architecture, the Evidence-based Agent Stack, underpins Insilica’s ToxIndex platform. The stack emphasizes traceability and reproducibility by requiring machine-actionable provenance and version-pinned data for every output.

The ToxIndex platform uses nine specialized agents to conduct structured toxicological analysis. These include agents for protocol definition, data retrieval across more than 2,000 databases, and extraction with explicit marking of missing data. Additional agents assess risk of bias, model causality, and quantify uncertainty before producing final recommendations.

The methodology aligns with regulatory principles such as TREAT and e-validation, embedding continuous monitoring and drift detection into the workflow. Insilica describes ToxIndex as the first operational implementation of evidence-based AI in regulatory toxicology.