January 2026
The FDA and EMA's joint principles call for human-centric design, data governance, and documentation across the AI life cycle.
fda.govEvidence infrastructure
Provenance
is non-negotiable.
The healthcare and life-sciences (HCLS) industry has stopped buying chatbots and started embedding AI agents into the workflows that move drugs forward: synthesizing research data, drafting clinical study reports and safety narratives, running root-cause investigations and batch-release decisions. The productivity case is real. But in every one of those workflows the agent produces an assertion that enters a regulated record or drives a high-consequence decision, and the agent cannot show its work.
Provenance is no longer optional.
July 2025
The EMA's draft Annex 22 would limit critical GMP AI to static, deterministic models.
ec.europa.euApril 2026
The FDA's first warning letter citing AI made ungoverned AI in a regulated record a citable violation.
fda.govThe problem
It's an architecture problem, not a model problem.
Put AI into a regulated workflow and it produces an assertion that enters a record or drives a decision, and it can't show its work. Accuracy is a property of a population of predictions. The evidence chain is a property of a single assertion, and you can't derive the second from the first.
The architecture
Reasoning on top, the record underneath, a gate between.
Generative AI does the reasoning. Every critical assertion beneath it is locked, provenanced and attested by a person. A gate between them decides what becomes part of the record. It's built on open standards (W3C PROV-O, SHACL and NGSI-LD), it federates across your existing systems, and it moves no data. Provenance travels across the site, CRO, sponsor and regulator boundaries that healthcare data legally can't cross.
Powered by the PNE
Evidence infrastructure is powered by the Provenance Neural Engine (PNE). At its heart, the PNE locks every critical assertion to its source, so AI outputs are verifiable by construction: an evidence chain from every AI decision back to source across the boundaries healthcare data can't cross.
Where it runs
Inside your boundary.
In your own cloud account or data center, as one deployable unit, on any cloud. The record is captured as data lands. Nothing phones home.
What stays yours
All of it.
The data, the record, the keys and the Institutional Brain built from them. We take no custody. Usage is reported by your own administrator.
What you get
An Institutional Brain.
Your own resolved, provenanced, queryable record of everything your systems and your AI have done. It compounds the longer it runs.
Every seat on the buying committee.
CFO Brief
Fundable by construction.
The CFO test: the five questions finance asks before it funds AI, and how evidence infrastructure passes each one.
Read the briefCMO Brief
Defensible by construction.
For clinical development: the AI-drafted submission that survives the audit, traceable to source.
Read the briefCAIO Brief
Governed by construction.
Deploy AI safely across all of GxP: generative reasoning over a deterministic, provenanced record.
Read the briefQuality & Regulatory Brief
Inspection-ready.
The audit trail you never rebuild by hand: Part 11, ALCOA+, and the Purolea precedent.
Read the briefCISO Brief
Secure by construction.
A control point, not new attack surface: Shadow AI, vendor AI, provenance as a security primitive.
Read the briefCDO Brief
Accountable by construction.
The agentic-governance mandate, held not rented: the CFO's controls for the agent stack.
Read the briefLet's build evidence infrastructure together.
Tell us the question your teams can't answer today. We will tell you whether the first release can.