
By Anita Modi, Co-Founder & CEO, Peer AI
Across industries right now, an important question is being asked about enterprise AI: when a system learns from the judgment, corrections, and reasoning of expert workers, who owns what it learned?
This question is being debated in financial services, in legal, in professional services. It hasn’t been asked loudly enough in our industry, pharmaceutical regulatory submissions, where the stakes are unusually high.
Last week, we wrote about why the judgment of the expert teams is the real edge when AI is used in a regulated industry like ours. Now we ask the next question: who owns it?
Most AI tools today let expert intelligence slip away
A senior regulatory writer working on a Biologics License Application (BLA) has spent years developing judgment that isn’t written down anywhere. She knows which safety signal FDA reviewers will probe on. She knows how to frame an ambiguous endpoint in a way that holds up in review. She knows the right language for presenting the results and handling data discrepancies. That accumulated knowledge, built across programs, agencies, and filing cycles, is arguably some of the most valuable institutional capital in drug development.
Most AI tools being adopted in regulatory writing today let that intelligence slip away. A writer uses a general-purpose model to draft a clinical study report (CSR) section. The writer gets faster at reviewing output she didn’t write. The organization captures efficiency. But her judgment, the reason she changed that paragraph, disappears the moment the document is final. Neither the writer nor the organization retains it.
Regulatory intelligence at the platform, organization and individual level?
Think about this on three different levels:
At the platform level: the durable advantage in AI isn’t the model, it’s the learning loop built on top. As foundation model capabilities converge, the advantage goes to sponsors whose workflows, traceability and quality controls sit on a platform where the learning stays theirs and compounds over time. Sponsors that simply adopt a model and call it a strategy will not have it.
At the organization level: the regulatory intelligence embedded in a submission workflow belongs to the sponsor and the writers who created it – not to the model vendor whose infrastructure it ran on. That means operating under zero data retention (ZDR) terms with model providers, and building traceability and quality control (QC) logic into the platform rather than into a foundation model’s training set. The traces a regulatory team leaves across years of filings are proprietary and should be handled that way.
At the individual level: a medical writer who spends years working with AI should leave with sharper judgment than when they started. From our perspective, that is a fundamental design choice. Sure – AI that generates output for writers to review makes writers faster. But AI that forces writers to set strategy, approve structure, and adjudicate every quality finding makes writers better. The second model cultivates expertise.
The organization owns the intelligence. The writer keeps the expertise.
How Peer AI keeps the intelligence with sponsors
At Peer AI, we’ve built our platform around this problem. Writers set the document strategy, approve the structure, and adjudicate every QC finding. The AI executes the work that doesn’t require expert judgment – mapping tables to sections, tracing every claim to source, running consistency checks across modules. We operate under ZDR terms with our model providers. The regulatory intelligence built in a Peer AI deployment stays with the organization and the writers who created it. That’s our design principle, from day one.
It’s an exciting time – pharma is moving fast on AI adoption. The question for every pharma leader is simple: in five years, will your best writers be sharper or just faster?
About the author
Anita Modi is the co-founder and CEO of Peer AI, the AI platform built for life sciences regulatory submissions.

