GitHub Copilot is now available for businesses, months after it was launched for educators and individual users.
The new plan is called GitHub Copilot for Business and costs $19 monthly. It includes all of the features available in the single-license Copilot tier, corporate licensing, and policy controls. IT administrators can block developers from seeing code that matches the public code on GitHub. This is likely to be a response to intellectual property controversies surrounding Copilot.
It’s powered by Codex, an AI model developed by OpenAI. Codex has been trained on billions upon billions of lines of code. It can suggest additional lines and functions in the context of existing codes. Copilot, which has over 400,000 subscribers, can present a programming solution or approach to developers in response to what they want to do (e.g., “Say hello to the world”) using its knowledge base and current context.
Users have prompted Copilot to generate code using Quake, code snippets from unique codebases, and example code from books such as “Mastering JavaScript” and “Think JavaScript.” GitHub admits that Copilot’s suggestions are about 1% longer than the training data.
GitHub asserts that fair use, the U.S. doctrine that allows the use of copyrighted materials without first obtaining permission from the rights holder, protects it in the case that Copilot was knowingly and unknowingly created against copyrighted codes. However, not all agree. In a class lawsuit, Microsoft, GitHub, and OpenAI are accused of violating copyright law by allowing Copilot permission to restate licensed code without credit.
Some legal experts argue that Copilot could pose a risk to companies if they unwittingly include copyrighted suggestions made by the tool in production software. Elaine Atwell writes in a Kolide blog piece that Copilot strips code from its licenses, so it is difficult to determine which code can be deployed and which may have conflicting terms.
GitHub has introduced a filter to its Copilot platform that compares code suggestions against their surrounding code. It hides suggestions that are not close matches or match the public GitHub code. Tim Davis, a Texas A&M University computer science professor, discovered that Copilot would emit large chunks of copyrighted codes, including all license text and attribution, by enabling the filter.
GitHub will introduce new features in 2023 to help developers make informed decisions about whether or not to use Copilot. These include identifying strings matching public code and referencing these repositories. GitHub Copilot Business customers will not be able to retain code snippets. This applies regardless of whether they come from public repositories (private repositories), non-GitHub repositories (local files), or other sources.
It’s still unclear if these steps will sufficiently quiet companies’ concerns about legal challenges.
Related Stories: