Who owns AI's work: In brief

Source lecture

The Science and Implications of Generative AI, Class 9 (Harvard Kennedy School, Spring 2024) (Harvard Kennedy School (Goel, Levy, Svoronos)). Watch on YouTube.

Lesson 7 takes one corner of the slow-change square that lesson 6 put on the risk map, the ownership question, and opens it into its own lesson. It is the copyright lesson, and it maps a fast-moving corner of the law as of July 2026, where cases and rules change often, sometimes month to month. It adapts the intellectual-property session of the Harvard Kennedy School course this track is built on, created by Sharad Goel, Dan Levy, and Teddy Svoronos, and it keeps that session’s calm structure. The promise is a map, not a verdict: a debate to understand so you can follow it in the news without being pushed around by either side. Because this is legal-adjacent ground, the lesson opens with a plain frame, it explains a public debate, it is not legal advice, and it does not tell you what is legal in your own situation.

Two questions arrive with anything you make using an AI tool. First, the model learned by studying millions of things other people wrote and drew, so were those people owed anything? Second, the new thing you shaped with your words, is it yours to own? The lesson answers each honestly, marking every legal claim as one of three kinds: a decided ruling from a single trial court with its limits stated, a settlement reported as a fact with no lesson drawn from it, or a genuinely open question left open on purpose with an “as of July 2026” hedge.

The capability: after this lesson, you can take a copyright headline about AI and place the claim in the right bucket, telling a court deciding one case on its own facts apart from settled law; you can state the four factors a fair-use question turns on; you can reason about which part of your own AI-assisted work is yours under the human-authorship split, using the Copyright Office’s registration practice rather than a rule you invent; and you can tell copying a specific work from imitating a general style, and name the separate laws, like the right of publicity, that may still apply.

What the lesson covers. First, why we protect ideas at all: copyright as a bargain that hands a creator a limited, temporary head start so people keep creating, after which the work enters the public domain. Then the training question, whether it was fair use to train a model on copyrighted work nobody paid for. This is where the most money rides and where courts have actually ruled since 2024: a Delaware court found one startup’s copying not fair use, and two California courts found a particular training use fair use while hedging hard, but the rulings are narrow, fact-specific, conflicting, and unreviewed by any higher court, so the general question stays genuinely open. The lesson keeps the anchor New York Times case at status level, reports the Anthropic settlement of at least 1.5 billion dollars as a fact with no merits inference, and lays out the four fair-use factors: the purpose of the use, the nature of the work, the amount used, and the effect on the market. Then it flips to your side of the screen: the bedrock rule of human authorship, why purely AI-made work cannot be copyrighted, and the practical split that your own creative selection, arrangement, and editing can be. It closes on the output question, the distinction that copying a specific work can infringe while imitating a general style generally does not, and the sharp open question of whether style should still be the line when a machine can imitate almost anyone at scale.

Why this order. The lesson runs from the shared bargain that explains every rule, out to the training fight, back to the reader’s own ownership, and finally to when an output copies too closely, so each answer builds on the same foundation rather than arriving as a loose fact. The bucketing runs throughout because the single most useful habit here is telling a ruling from a rule: one court deciding one case on its own facts, with an appeal pending, is a data point, not a nationwide verdict. The practice runs a bucketing drill and a reflection on owning your own AI-assisted work, with an optional plain conversation in Clawless, and it repeats the education-not-legal-advice frame before the task. Lesson 8 turns from who owns the work to who does the work, the most personal question in this whole track for many readers, met with the same care.