Fix HeinOnline with AI access

Law libraries should be able to buy permission for AI-assisted research, with stated limits and rules that protect the collection.

I want to give a very bright research assistant an assignment: find the law review literature before 1960 on a liability insurer’s duty to settle, identify six or eight articles worth reading, and pull the passages discussing bad-faith exposure after a refused settlement. Pay particular attention to predictions about the economic effects, including the price of insurance. Come back with a table showing the citation, the relevant passage, the page number, and why the article belongs in the pile. I would then read the passages, check the sources, and decide what to pursue.

This is an ordinary faculty research task, well suited to HeinOnline, a subscription legal research database with a deep journal archive. It involves searching, opening articles, following citations, and discarding things that looked promising until someone read them. Delegating this preliminary work lets me pursue other parts of the project and benefit from what the assistant learns along the way.

The assistant I have in mind, however, is a large language model. With all due respect to student research assistants, I suspect it could do this work better and faster. It might eventually outperform even a skilled research librarian working without AI, although that is a higher bar. These days, my choice would likely be GPT-6 Astra, which OpenAI describes as capable of computer use and multistep research. (Astra helped me draft this post, by the way; so too Claude Fable.) But another model capable of operating a browser could do the job. I want to log in through my university, open HeinOnline, and ask the assistant to work in that session, poking at the HeinOnline website until it intelligently yields the desired information.

I have not run that experiment, and I do not intend to do so in the immediate future. The obstacle is Hein’s published terms of service, which prohibit the workflow. Saving myself an afternoon or gratifying an urge to explore all things AI is not the best reason to create an access dispute for my library or law school. But I believe those restrictions are overbroad and that relaxing them presents a business opportunity for Hein. I hope this post prompts a serious discussion of both.

Why Hein?

Hein pairs an extraordinary archive with an interface that makes me feel as though I have returned to 2010. Its journal collection is unusually deep, often running back to a title’s first issue, and its searchable images preserve the original pages. For the historical project I have described, I want that collection available to my assistant in one place.

Hein’s Terms and Conditions appear to prohibit both parts of the task I want to delegate: having an agent retrieve articles and having an AI tool analyze them. The restriction reaches even a professor who downloads one article herself and uploads it to Claude for a summary. I think Hein should welcome that use, provided the article is kept private and the AI provider cannot retain it for training. Someone is reading the scholarship its service helps them find.

The terms contemplate the possibility of a written agreement to allow for automated access, but I do not plan to request an individual exception. I expect that would be a waste of time. Hein is unlikely to change its terms or technical arrangements for one professor, even one with the very best intentions. The change I want is permission available to subscribers generally, or something libraries can negotiate as part of their subscriptions.

Hein has legitimate reasons to impose conditions on that permission. An agent could overwhelm its servers or collect enough articles to help someone assemble a competing database. An outside AI provider might retain uploaded text, use it for training, or make it available to others. None of those concerns disappears because the software is acting for a professor. But reading six articles need not produce any of those results. The agreement should specify the conditions under which that research may proceed.

Hein’s own use of AI provides a starting point for defining those conditions. It already generates article summaries. According to its AI FAQ, that work occurs within its internal network, and its models do not learn from user activity. Hein can control that arrangement more directly than it can control a professor’s uploads to an outside service. It could nevertheless permit outside tools whose terms and settings provide specified protections.

The safeguards should address each risk directly. A limit on requests per minute can protect the servers. Limits on cumulative retrieval, combined with a prohibition on pooling accounts to evade them, can make large-scale extraction harder. Binding provider commitments can restrict retention, training, and disclosure. Hein’s logs can help establish whether I exceeded a retrieval limit; they cannot establish what a provider later did with an uploaded article.

Better terms

Hein should give subscribers express permission to search, retrieve, and analyze articles with AI, subject to limits they can understand before they begin. Subscribers who stay within those limits should be able to get on with their research. Here is a sketch of three provisions:

Permission. An Authorized User may direct an Agent to search, retrieve, display, download, and analyze Materials for purposes permitted by the institution’s subscription. This permission controls over conflicting restrictions on automation, AI processing, compilation, and transmission to approved processors. No separate permission is required for each task within the agreed allowance.

Counting. The subscription’s Rate Schedule must state limits on retrieval requests, distinct documents, and distinct source pages, together with the applicable measurement periods. Viewing, downloading, and text delivery count alike. Repeated retrieval counts against request frequency but counts each document and page only once within the applicable volume period.

Reaching a limit. Reaching an allowance pauses further retrieval until capacity becomes available; it is not, by itself, a breach. The Agent must obey the pause. A User does not breach this agreement merely because other Users have exhausted the institution’s allowance.

Hein should supply the numbers because it knows more about its server capacity and subscription economics than I do. The schedule should address both bursts of requests and cumulative retrieval over longer periods. Users should be able to see how much of their allowance remains and when it resets. A researcher with a larger project could obtain a larger allowance at a stated price.

Some details

1. Putting these terms into practice may require some technical changes. Hein’s existing 200-page download limit applies to individual downloads: its help page tells users to download longer documents in successive 200-page intervals. Its public usage portal describes institutional statistics. Neither establishes that Hein already meters cumulative retrieval by individual professors. Hein may need user-specific identifiers, additional counters, and interface changes. I suspect those costs would be small relative to the improvement in the research experience, though Hein would have to assess them.
2. The permission must include the ability to develop offline collections of Hein-originated material within stated limits. It's essential to be able to collaborate with AI in exploring a compilation of data. This includes all the activities of which AI is capable: building websites, creating presentations, creating structured data. Just not in a way that is more than fair use of the underlying materials. Also, accessibility accommodations should preserve equivalent access without authorizing unlimited extraction.
3. To deter abuse, Hein can certainly retain remedies that let it stop an agent that evades limits or disrupts service, with notice of the breach and a chance to cure. But one professor’s conduct should not cost the university its subscription.

The business case

Giving AI agents access to Hein would not be an act of charity on its part. On the contrary, permitting assisted research would let Hein earn revenue from a use its current terms prohibit. An institution could buy a retrieval allowance suited to its researchers and pay for additional capacity when needed. Hein would retain controls over retrieval and impose express restrictions on training, retention, and redistribution. Hein's expenses at the front end would be small. Although it certainly could modify its website to be friendlier to agentic browsing, the agents have probably gotten smart enough that they don't need Hein to do so. The major expense would be extra use of its servers, which to the extent Hein does not have excess capacity, could be something for which Hein could charge as part of an AI-friendly subscription.

Assisted access could also strengthen the case for renewing the underlying subscription by helping researchers get more useful work done. Increased traffic alone would establish little: Hein’s usage definitions count repeated requests for the same article, so an agent stuck reopening a PDF could inflate a usage report. The stronger renewal argument is that researchers can find relevant sources, check passages, and complete projects through the tools they prefer.

Keeping the prohibition creates the opposite risk: researchers may increasingly rely on sources that permit their preferred workflow. Law Review Commons already offers more than 300 open-access law reviews, including historical archives. It does not duplicate Hein’s combined collection, and open access alone does not authorize automation. But wherever another source permits assisted research, scholars have a reason to start there. Repeated success elsewhere could weaken their institution’s reason to keep paying for Hein. It could also create an impetus for "Uberification" of legal research, with capitalized entities going in to various law reviews, reaching agreements with them to permit AI-based access, and then creating a comprehensive modern website based on the scraped materials. With advances in AI coding abilities, that last part of the project will no longer be particularly difficult.

To capture the business contemplated here, Hein would need to secure any additional publisher permissions and provide a supported means of access. It should amend agreements where necessary and identify collections subject to different conditions. A paid API or connector could make retrieval easier to meter and support. Whichever interface Hein supports, subscribers should receive express permission to retrieve and analyze the material through it.

Legal scholarship

For legal scholars, the benefit of an AI-friendly Hein would be a capable research assistant able to work with a collection their libraries already pay for. Like a human assistant, it could find articles, follow citations, compare arguments, and pull passages for closer examination. That could expand what a scholar can investigate and produce. Hein should make that work possible under clear terms.

Delegating research to AI need not cause any more “brain rot” than delegating it to another person. That is a larger debate, but we need to get past the idea that using AI somehow taints legal scholarship. I see nothing wrong with using it to search, analyze, or draft. The questions are whether the resulting work is accurate, whether its arguments hold up, and whether it teaches us something worth knowing. A human author’s labor does not make an unsupported claim sound, and AI assistance does not make a sound argument defective. Scholars remain responsible for checking sources, representing others’ work fairly, and supporting what they publish. Their jobs, however, can start to resemble that of a "research and publication partner" rather than those of a forever associate stuck slowly repeating rote workflows over narrower domains.

The purpose of legal scholarship is to enlighten: to help us understand the law, identify its failures, and make better decisions about it. Faculty hiring and promotion have also made scholarship a way to rank people by their individual research and writing abilities. Those purposes can come apart as AI becomes better at the work. Preserving a competition in unaided performance, for whatever value that may occasionally provide, is a poor reason to limit the knowledge we produce.

If AI can help us produce better scholarship faster, we should use it. If it can perform parts of that work better than we can, we should let it. The contribution should be judged by what readers learn and what they can do with it. Changing Hein’s terms would remove one concrete obstacle to producing better and more comprehensive legal scholarship.