Executable Knowledge Artifacts and the Law: The Problem and a Framework

For most of professional history, experience could not be separated from the professional who acquired it. Forms, manuals, and software captured portions of professional judgment, but the judgment itself remained human. That practical limitation preserved two distinctions the law depends on: the distinction between an employer’s property and an employee’s knowledge and skill, and the distinction between one client’s confidences and the next client’s matter. However, modern artificial intelligence (AI) removes this limitation. A professional can now reduce years of judgment to a machine-usable artifact that performs part of the work without the professional, and an artifact of this kind makes professional knowledge persistent, transferable, and usable apart from the person who developed it. Machine-useable artifacts can include skills, routines, playbooks, etc., built using AI, such as a SKILL.MD for Claude or ChatGPT.

Consider an associate at a patent firm who spends six years prosecuting applications for a single client. The associate learns which claim formats the client prefers, which gaps recur in the invention disclosures the client submits, how the client weighs continuation costs against claim scope, and how to effectively target competitors or cover the client’s own products. The associate reduces that judgment to an executable knowledge artifact, refines it over two years of additional work, and then leaves for another firm. The artifact contains no client document and copies no client language, but it performs part of the work the associate was hired to do. Suddenly it is unclear who may use the artifact, and importantly who may prevent its use.

Now change one fact. The associate does not leave, the artifact is promoted to the firm’s shared library, and a lawyer in another office invokes it while prosecuting an application for a different client. No one retrieved a form from the original client’s file, and no one screened against the original client’s matter, because no lawyer who worked on that matter touched the new one. The question is no longer who owns the artifact. It is whether the firm has used one client’s confidential information in another client’s matter via indirect knowledge transfer, and whether anyone at the firm is in a position to know that it did.

Both problems are traditionally discussed as questions of ownership, but ownership is incomplete at best to answer these questions. A single artifact can reflect an employee’s experience, an employer’s investment, knowledge developed in a client’s matter, and material drawn from public sources. Determining who may use, copy, transfer, disclose, or retain the knowledge the artifact carries is more complicated than traditional legal concepts provide. What follows separates the artifact from the rights in the knowledge the artifact holds.

I. What Has Changed

Externalized professional experience is not new. Form files, model claim sets, and brief banks are all professional judgment reduced to a document, and the profession resolved the question of their use long ago: a lawyer may reuse them, provided client-specific material is removed. Two things about executable knowledge artifacts are different.

The first is that the rules governing a lawyer’s knowledge assume the lawyer can do nothing about that knowledge. A lawyer who has represented one client for six years cannot unlearn what the representation taught her, cannot separate it from what she applies to the next client, and cannot produce it on demand. The rules accommodate that condition by regulating what she does with the knowledge rather than where she keeps it. The American Bar Association’s ABA Model Rules of Professional Conduct, at Model Rule 1.9(c), prohibits using information relating to a former client’s representation to the disadvantage of that client, but it says nothing about storage, because the lawyer has no say in where the information is stored (in her mind). An executable knowledge artifact, on the other hand, can be inspected, copied, produced, and deleted, and a rule addressed only to use was sensible only while storage was involuntary.

The second is that the artifact can be applied forward. A brief bank is inert. Someone must remember that it exists, retrieve the right document, and conclude that the present matter is one the document fits, and that act of retrieval is where the judgment about client-specific content was exercised. A lawyer who pulls a set of claims from a prior matter knows whose matter it was and edits accordingly. An executable knowledge artifact is invoked because the present task resembles the task the artifact was written for, and the tool acts on new facts rather than merely providing static guidance.

II. What the Artifact Contains

An executable knowledge artifact is a mixture. The content of a single artifact can come from at least four sources, and each source is protected, if it is protected at all, by a different body of law.

• Confidences and information specific to a client’s matter, including the client’s technology, its unpublished applications, its prosecution strategy, and its tolerance for cost.
• Work product and information belonging to the employer, including material developed at the firm’s expense and information that may qualify as a trade secret.
• The individual attorney’s general knowledge, skill, judgment, and experience, which the law has long treated as the employee’s own and which an employer generally may not prevent the employee from using.
• Public material, including statutes, rules, published decisions, published applications, and the ordinary technique of the field.

The four sources cannot be told apart by reading the artifact. An instruction directing that a particular structural limitation be avoided in every application may be ordinary drafting practice, a lesson learned at the firm’s expense, or a rule derived entirely from one client’s portfolio and the litigation behind it. The text is the same in all three cases. The lawyer who wrote the instruction may be the only person who can say which it is, and she may be able to say so only at the moment she writes it, if at all. Or in an even more ambiguous situation, an AI may have derived a preference or style based on a pattern or rule that humans never identified for a particular client.

III. What Interest Is Being Asserted

Entitlement to the artifact is the question asked most often and the least useful place to begin, because title to a thing does not determine who may use the knowledge the thing contains. At least six interests should be identified separately.

• Ownership: who holds title to the artifact as property.
• Use: who may invoke the artifact, and in which matters.
• Copying: who may reproduce the artifact or write another artifact from it.
• Portability: whether the artifact, or the knowledge in it, may leave with the person who wrote it.
• Disclosure: who may be told what the artifact contains, including in discovery, and who must be told that it exists.
• Access and retention: who may reach the artifact within the firm, and whether the artifact must be preserved or destroyed.

These interests do not travel together, and no party needs to hold all of them. A firm may own an artifact that it may not use in a particular matter. A departing lawyer may be entitled to the knowledge in an artifact without being entitled to the file that contains it. Resolving one interest resolves none of the others, and a dispute framed as a dispute over ownership will produce an answer that satisfies no one.

IV. What the Existing Law Says

The employer’s investment is protected, if at all, by trade secret law and by the employment relationship. Trade secret law requires the employer to have taken reasonable measures to keep the information secret and prohibits the employee from misappropriating it, but it excludes the employee’s general knowledge, skill, and experience from protection. That exclusion is the difficulty, because an executable knowledge artifact is general knowledge, skill, and experience written down. The professional conduct authorities send the question back to the same place. ABA Formal Opinion 99-414 states that a departing lawyer does not violate any Model Rule by taking with her copies of documents she has created for general use in her practice, and it directs what remains to “principles of property law and trade secret law.” ABA Formal Opinion 489 addresses lawyer departures at length and never mentions work product. Property law asks who owns the document, and trade secret law asks whether the firm protected it. Neither asks whether the knowledge in the document came from a client’s matter, which is the interest most likely to be harmed and the one interest the lawyer cannot settle by negotiating with her firm.

The client’s interest is governed by the rules of professional conduct, which were written on the assumption that the information is held in a lawyer’s memory or in a client’s file. Model Rules 1.6 and 1.9(c) prohibit disclosing information relating to a representation and using it to a former client’s disadvantage. Model Rule 1.16(d) requires the lawyer to surrender papers and property to which the client is entitled at the end of a representation, and it applies to electronic data, so whatever storage design a firm adopts, the firm should be able to separate one client’s material from the rest and hand it over. However, the firm may not be obligated keep the material apart in the first place. The cloud-computing opinions come closest to answering this issue, and they require reasonable care rather than isolation. Commingling, the term the profession uses for a failure to keep things apart, is limited to trust accounting and has not been applied to data.

One authority addresses the situation in which information from one client’s matter reappears in another client’s matter. ABA Formal Opinion 512, issued in 2024, observes that self-learning tools raise the risk that information relating to one client’s representation will be disclosed improperly, and later revealed in response to prompts by lawyers working on other matters. What the opinion requires in response is the client’s informed consent. It does not address how the information came to be in the tool, and it does not say what a firm should do about an artifact that already exists.

V. Experience Now Has Metadata

Every question identified above is a question of fact, and until recently those facts could be proved only by testimony and by inference drawn from the lawyer’s later work. An executable knowledge artifact carries its own record. Authorship, revision history, source materials, evaluation results, and use logs can show when a piece of judgment was added, how it was changed, what it was derived from, and where it was applied. Human experience rarely carries a record of that kind. A professional may remember when a lesson was learned, but memory does not produce a version history. Continued development and use generate more of the record rather than less. Even experience now has metadata.

The record does not necessarily favor the party who asks for it. It may show that a piece of judgment predated the employment or came from a source in which the employer invested nothing. It may also show that an artifact written in one client’s matter was invoked in another’s, on a particular date, by a named lawyer. A firm that generates a record of this kind should expect to produce it in discovery.

A proposed solution will be the subject of a future article. Keeping one client’s knowledge apart from another’s is a condition that a client cannot audit and a firm cannot prove, while a record of where the knowledge came from is something a client can demand, a firm can produce, and a lawyer can certify. No party is well served by a promise of separation, and every party is served by a record of origin, which is why recommendations written for the client, for the departing lawyer, and for the firm converge on the same practices even though the three parties want different things from them.

Image Source: Deposit Photos
Image ID: 672857744
Author: supatman