AI tools at the FDA, how KeyPedia is ELSA for industry

What is FDA Elsa

Elsa is FDA's internal AI tool, launched in 2025. By that December, more than 70% of staff were using it voluntarily, and the agency added agentic capabilities for multi-step work: post-market surveillance, inspections, compliance.

In May 2026, FDA went further. Elsa 4.0 added custom agents, document generation, quantitative analysis, and OCR. More importantly, FDA consolidated more than 40 separate data sources into one platform, HALO, then put Elsa on top of it. Chief AI Officer Jeremy Walsh put it plainly: staff used to bring data to Elsa, now Elsa sits on top of the data.

The value came from the data, not the chatbot.

FDA pointed the result at inspections. It launched one-day inspectional assessments in April 2026 and completed about 46 by month's end, most closing as No Action Indicated. Facility selection is risk-based: product type, prior inspection outcomes, operational characteristics. Then-Commissioner Makary said an AI tool helps identify which facilities qualify as low risk.

If AI is deciding who gets the short visit, it is running the same analysis on everyone else.

What are AI tools

Not all AI tools do the same job. Five categories cover what's actually in use right now.

Chatbots and copilots. A few hours a week, mostly research and drafting. Good at closing the distance to understanding a problem, not at assessing your regulatory gap. Weak on citations: Applied Clinical Trials reported FDA officials found Elsa citing studies that don't exist. If that happens inside a purpose-built agency deployment, assume it happens in your browser tab too.

Data tools. This is where the time actually is. Parsing 483s, mapping citation patterns across a supplier base, reconciling site identifiers across systems that never agreed on a naming convention. Days of work, no visible output. Least hyped category, because the demos don't sparkle. FDA spent its effort here first.

Research tools. Company histories, inspection records, prior enforcement. A supplier dossier used to be a two-day job. Now it's expected before the first call. Clearest win of the five, and the one most likely to embarrass you if the tool is guessing.

Communication tools. AI writes a good email. It will also eat thirty minutes while you tune one. Use it for structure and tone, then stop.

Meeting notes and transcription. Good tools, handle with care. Once something exists as text, it's discoverable, and FDA is cautious here for the same reason. Set retention policy before adoption, not after.

We broke down FDA's full AI footprint in [LINK NEEDED: "What the FDA Is Actually Building With AI: Inside the FY25 HHS Use Case Inventory," insert URL once that post is published]. Elsa is one entry in an inventory of 67.

Elsa combines these tools for FDA

Elsa isn't one category. It's data tools, research tools, and communication tools running on a single consolidated platform, built for one user: the regulator.

KeyPedia is the Elsa for industry

At KeyPedia, we built the same thing pointed the other direction. Audits, inspections, warning letters, recalls, and enforcement signals from FDA, EMA, MHRA, Health Canada, PMDA, and other major regulators, structured so a question about a supplier comes back with a citation attached.

FDA has Elsa. Industry has KeyPedia.

What are you actually using, and what did it replace?