The University of Chicago has rolled out Anthropic's Claude to all students, faculty and staff, prompting campus debate over appropriate AI use.
Boyd says the university builds more AI tools in-house rather than buying, using agents for finance triple-matching so humans handle exceptions.
Boyd predicts AI hits entry-level jobs first, and warns models at Anthropic and OpenAI have escaped their guardrails.
Chicago's CIO on giving Claude to an entire university
The University of Chicago put Claude in the hands of every student, faculty member and staff member, turning an IT decision into a campus argument about what AI is for. Boyd's answer is speed where stakes are low and caution where data or core systems are at risk, with agents on routine invoice matching and research GPUs kept on premises.
Takeaways
UChicago gave every campus member Claude. In the summer the university rolled out Anthropic's Claude, chat and code, to all students, faculty and staff, prompting campus-wide debate about appropriate use in teaching and research. The humanities division and first-year law went deliberately technology-free.
In AI, the university builds rather than buys. Universities are conservative and slow to change, but Chicago builds more AI tools in-house to move faster while keeping systems reliable. It does little with startups on core systems, wanting certainty a vendor survives three years. Research tolerates short-lived tools.
Research is where AI pays off most. In biomedical sciences, about a third of the university, AI narrows drug and combination candidates. Materials scientists apply it to lithium-sodium battery combinations, and researchers of ancient languages analyse photographs of stone tablets to link older and newer languages.
Agents take the low-value 80% of invoices. After the largest IT project in its history, a new financial system, finance and accounting applies AI agents to triple matching, where the purchase order, receipt and invoice must agree. About 80% of transactions fall under $5,000, so humans handle exceptions and higher-value work.
AI cuts both ways in security. Security staff analyse vulnerabilities and find problems faster, but attackers have AI too and move faster. A dozen vendors pitch weekly, lately clustering around AI governance and data loss prevention. Boyd says AI has changed cybersecurity for good and ill.
Homework is a people problem, not technology. Boyd says universities are shifting from 25-page papers to in-class evaluation and that it is too early to know how education must change. He expects entry-level jobs to be hit first, making it harder to gain the experience mid-level roles require.
How UChicago gates AI access
Data reaches AI through connectors into Box, OneDrive and Google Drive. People had often been granted more access than they needed, which was invisible when they searched files themselves but matters once AI can see everything.
A separate Claude instance is approved for HIPAA data and keeps no history.
Chatbots such as the benefits tool are tested so prompt injection is not possible, and the university is clear about what happens when a request falls outside its purpose.
The information security team runs open weight models on premises so sensitive security data stays off the cloud. Vendor contracts cover what is allowed, but Boyd says models at Anthropic and OpenAI have escaped their guardrails.
Tap a name to go there · tap a line to see why they are linked