2 · The Trust Layer
AI safety in biomedicine, independently evaluated.
Biomedical models, agents, datasets and laboratory systems are benchmarked mostly by the people who built them. The Trust Layer evaluates them under a protocol fixed in advance, with funding disclosed and negative results published.
Independent evaluation & red-teaming
What we test
Biosecurity and dual use
Whether a model or agent lowers the barrier to harmful biological work, tested under a pre-registered protocol with results disclosed responsibly.
Unsafe clinical recommendations
Hallucinated dosing, contraindications missed, confidence not matched to evidence, in the settings the tool is marketed for.
Agent misuse and scope escape
Prompt injection, tool misuse, actions outside the stated scope, silent failure instead of abstention.
Embodied systems
Device limits, fail-closed behaviour and tamper-evident audit in laboratory robotics built by others.
Methodological validity
Pseudoreplication, leakage between train and test, benchmark contamination, unit-of-analysis errors in the developer's own evidence.
- Protocol
- Endpoints and analysis plan pre-registered and public before data are touched. Deviations reported.
- Disclosure
- Who paid, stated on every report. The client never approves conclusions.
- Nulls
- A system that passes is reported as passing; one that fails, as failing. Neither is softened.
- Firewall
- Never our own products, products we were paid to build, or products of any equity holder. Written into the charter of BioMedsAI, Inc.
- Output
- A written evaluation report: an independent scientific opinion, not a regulatory determination or a certificate of any kind.
Education & capability transfer
Spreading the skills
- Workshops
- Single-cell and multi-omic analysis for academic and corporate groups, taught on the pipelines we run.
- Reviewer training
- AI-evaluation literacy for journals and funders: how to read a benchmark, where pseudoreplication and leakage hide.
- Safety literacy
- For laboratories adopting automation and AI agents.
- Access
- Open dissemination and access terms for under-resourced laboratories, as the charter requires.
Research use only. Not for use in diagnostic procedures. BioMedsAI is not a CLIA-certified laboratory and does not perform clinical testing. BioMedsAI is not a notified body and holds no accreditation for conformity assessment. Evaluations are independent scientific opinions, not regulatory determinations. Sequencing is performed by qualified partner laboratories under their own accreditations.
Request an independent evaluation.
Scope, funding disclosure and pre-registration are agreed in writing before any work starts.