What can the law teach us about where AI s ystems should exercise judgment when their instructions leave room for inte rpretation?
AI developers now govern their most capable models throug h written “constitutions.” These documents set out a model’s values\, prior ities and character\, and they’re used directly in training. Anthropic’s co nstitution for Claude is the most prominent example. These documents raise questions that law has dealt with for centuries. Who has the authority to w rite them? How should vague principles be interpreted when they conflict? C an a model meaningfully consent to the values it’s trained on? What role\, if any\, should governments play?
Nicholas Caputo (Johns Hopkins Scho ol of Government and Policy\; Oxford Martin AI Governance Initiative) and J ordi Weinstock (Harvard Law School\; Berkman Klein Center) will discuss wha t legal theory can and can’t tell us about AI constitutions. Caputo’s resea rch treats AI alignment as a problem of jurisprudence\, and his work in pro gress asks whether Claude can consent to its own constitution. Weinstock te aches at Harvard Law School about the confluence of AI and traditional law and served as an external reviewer of Claude’s constitution.
CATEGORIES:Speaker/Panel LOCATION:Lewis Hall\, 5th floor at the Berkman Klein Center's Multi-Purpose Room 515 GEO:0.000000;0.000000 ORGANIZER;CN="Matt Kristoffersen":MAILTO:mkristoffersen@jd27.law.harvard.ed u URL;VALUE=URI:https://hls.harvard.edu/events/how-to-tell-an-ai-what-to-do-r ules-cases-and-interpretation-2/ ATTACH;FMTTYPE=image/jpeg:https://hls.harvard.edu/wp-content/uploads/2026/10/event-thumbnail.jpg END:VEVENT BEGIN:VTIMEZONE TZID:America/New_York BEGIN:STANDARD TZOFFSETFROM:-0400 TZOFFSETTO:-0500 DTSTART:20261101T060000 TZNAME:EST END:STANDARD END:VTIMEZONE END:VCALENDAR