GitHub ships a code-generating assistant, and the inventorship and disclosure questions arrive with it

2021-06-29In-House

Nat Friedman, "Introducing GitHub Copilot: your AI pair programmer," github.blog (June 29, 2021)

GitHub announced a technical preview of GitHub Copilot, which it describes as an artificial intelligence pair programmer. The announcement states that GitHub Copilot was developed in collaboration with OpenAI and is powered by OpenAI Codex, a new system created by OpenAI that was trained on a data set that includes a much larger concentration of public source code than earlier models. The tool draws context from the code the developer is working on and suggests whole lines or entire functions inside the editor, and the announcement says the preview works especially well for Python, JavaScript, TypeScript, Ruby and Go while working with a broad set of frameworks and languages. The announcement says nothing about who owns a suggested function, nothing about whether a suggestion may reproduce code from the training set verbatim, nothing about the license terms attached to the public code the model was trained on, and nothing about where the developer's own code is sent in order to produce a suggestion. Those omissions matter to a patent program because each one maps onto a legal question that the Patent Act answers for people and not for tools: who conceived, whether the applicant owns what was conceived, and whether the invention was disclosed to a third party before an application was filed.

What it changesThe in-house program needs a written position before engineers adopt the tool. First, inventorship: Section 100(f) of Title 35 of the United States Code defines an inventor as an individual, so a function that the tool produced in response to a comment has no inventor unless a person conceived the solution the function implements. Disclosure forms should ask whether generated code was used and what the engineer specified. Second, ownership: code generated from a model trained on public repositories can resemble licensed code, and the assignment chain for a patent application should not rest on a block nobody at the company wrote. Third, confidential disclosure: the tool sends the surrounding code to a third-party service, so the vendor terms should be read for confidentiality and no-training clauses before any pre-filing invention is typed into an editor with the tool on. A policy that permits the tool for boilerplate and requires a human-authored record for anything in a disclosure keeps the filing clean.

More on Inventorship and AI-assisted invention · Running an in-house patent program

← All recent developments · Search all developments

Burbage Law is the practice of Jon Burbage, a registered patent attorney. 342 United States patents have issued from applications he wrote; the full list is on the Patents page. More about Jon Burbage.

This is general information about a published decision, agency guidance or piece of legislation. It is not legal advice and does not create an attorney-client relationship. Items concerning offices outside the United States are reported for information only and are not counsel on the law of those countries.