AI Tool Training

In July 2026, Harvard Medical School IT began offering an approved AI tool to every eligible employee and student. Before anyone could get access, they needed to understand which data each tool could safely hold. I researched, designed, and built AI Tool Training, the required course that turns that policy into decisions that people can act on. Completing it activates a Harvard AI Sandbox account and opens the request form for Claude, ChatGPT Edu, or Microsoft Copilot Premium.

The Challenge

The rules that learners needed were spread across several HMS and Harvard pages, written for compliance rather than for daily work. Learners, meanwhile, came for one thing: access to a tool. The course had to make policy feel useful instead of like a hurdle, and send each person into the right tool ready to use it well.

Design Decisions

• Training as the door to access. The course is the step that unlocks the tools, so learning sits at the moment when people are most motivated, not in a reminder email weeks later.
• Policy framed as questions that learners actually ask. Instead of restating Harvard’s data security levels, the course asks “What can I put in this tool?” Learners see that most approved tools can hold Level 3 data and below, that Level 4 and 5 data need a consultation with HMS Risk Management, and why public AI tools can’t be used for HMS work at all.
• Tool choice treated as a matching problem. A comparison tool helps learners pick Claude, ChatGPT Edu, or Copilot Premium based on the work that they do, so the request they submit is the one that fits.
• Access followed by value. A lesson on writing prompts around task, context, and expectation, plus guidance on checking and citing AI output, means that people leave able to do good work, not just allowed to.

Choosing a tool

Course diagram

Learners get the Harvard AI Sandbox for comparing models, plus one dedicated premium tool. The diagram makes the “and, then one of” rule clear at a glance.

Prompting with intent

Course diagram

The prompting lesson reduces good prompting to three ingredients, then shows one prompt improving as each is added.

My Role

• Researched HMS and Harvard AI guidelines and translated them into learner-facing content
• Designed the learning path and wrote every lesson
• Built the course in Canvas, including the tool comparison
• Worked with HMS IT colleagues on eligibility, the access process, and launch communications

Impact

• Serves as the required first step for every eligible HMS employee and student who uses the School’s approved AI tools
• Made HMS IT’s July 2026 launch possible by giving every learner the same foundation before access
• Prepared learners for the HMS AI Acceptable Use Policy, effective October 28, 2026, whose core rules the course already taught: use only approved tools, keep data at or below each tool’s level, and verify and disclose AI output

Built to Last

• Each tool has its own lesson, so HMS IT can add or retire a tool without rebuilding the course
• Policy content points to the published HMS IT list of approved tools and data levels, so the course stays current when that list changes
• Learners who need a tool that isn’t approved are pointed to the Vendor Assessment Request, so the course offers a next step instead of a dead end

Program and policy details are public from HMS IT: Access Claude, ChatGPT Edu, and Microsoft Copilot (opens in new tab), Generative AI guidelines (opens in new tab), HMS AI Acceptable Use Policy (opens in new tab), and Get an approved AI tool for your HMS work (opens in new tab).