Both under Oracle, OpenJDK and GraalVM hold opposing stances on contributions to AI code.

Two related projects supported by Oracle have issued opposite policies regarding open - source contributions created using generative AI. The OpenJDK Management Committee has approved a temporary policy that prohibits such contributions, while the Coding Assistants policy of GraalVM allows them. Both projects require contributors to sign the same Oracle Contributor Agreement (OCA) to handle intellectual property - related matters.

In early April 2026, OpenJDK released its policy widely banning generative AI content:

Contributions must not contain content partially or fully generated by large language models, diffusion models, or similar deep - learning systems. The content referred to here includes, but is not limited to, source code, text, and images in the OpenJDK Git repository, GitHub pull requests, emails, wiki pages, and JBS issues.

The policy gives three reasons. The first is the review burden: a large amount of seemingly reasonable but actually incorrect or difficult - to - maintain code will consume limited review time. The second is security and safety assurance: the JDK supports critical - mission systems, so a high threshold must be maintained. The third is intellectual property: the OCA requires contributors to own the intellectual property they grant to Oracle without restrictions, but there are still ongoing lawsuits regarding whether individuals own the intellectual property of AI - generated output.

The policy leaves room for private use. Contributors can still use generative AI to understand, debug, and review OpenJDK code, and also for research. They just cannot contribute AI - generated content. The policy FAQ clearly states that even if only 10 out of 100 lines of AI - generated code are modified, the contribution cannot be submitted because it still contains AI - generated content. The policy also allows the use of tools such as "spell - checking, grammar - checking, auto - completion, and refactoring functions in editors or IDEs" as long as "they are not based on large language models or similar deep - learning systems".

OpenJDK contributors will soon have to check a checkbox in the Skara automated pull - request review system to confirm that their contributions comply with the generative AI policy. OpenJDK acknowledges that it is often difficult to distinguish between human - generated and AI - generated content, but still recommends that reviewers watch out for typical traces of AI - generated content.

In mid - April 2026, GraalVM, an Oracle Labs project not under the jurisdiction of the OpenJDK Management Committee, clarified its AI - assisted contribution policy and contributor guidelines, allowing generative AI content:

GraalVM contributors can use AI coding assistants and similar tools when preparing contributions. … For the purposes of this document, "coding assistants" include AI tools that help draft, transform, interpret, review, or summarize code, tests, documentation, or submission text. This policy applies to contributions and project interactions prepared using such tools, including pull requests and issues submitted to the project.

The project also added a "Documentation Terms and Style Guide" for AI coding assistants on June 3, 2026.

The GraalVM policy references the AI Coding Assistants policy of the Linux kernel but also makes adjustments. For example, the Linux policy states that "contributions should include an Assisted - by label". In contrast, GraalVM's requirements are relatively lenient. It says that contributors can choose whether to specify which model or tool was used; however, if disclosing the AI - assisted process helps reviewers understand how the change was made, the project still encourages contributors to explain.

Contributor responsibility is at the core of GraalVM's contributor responsibility rules: the human contributor submitting the contribution is responsible for the entire contribution, including any part completed with AI assistance. They must review and understand the contribution, verify its correctness, and answer reviewers' questions without shifting the blame to the tools: "If a contributor cannot explain, defend, or maintain an AI - assisted change, the contribution may be rejected."

The reviewers' rules for maintainers will not change due to AI assistance. GraalVM clearly states that using AI assistance does not mean that a change is correct, nor does it mean that it is ready for review, and it definitely does not mean that it can skip the normal review process. When reviewing, maintainers can still ask about the source of the change, design intent, licensing, testing, and whether the contributor truly understands the change.

Both projects require contributors to sign the same OCA and grant Oracle unrestricted intellectual property rights. However, OpenJDK uses the unclear intellectual property risks brought by AI - generated content as a reason for a full - scale ban, while GraalVM believes that contributor responsibility is sufficient to allow such contributions.

Oracle is formulating a comprehensive AI contribution policy for OpenJDK and will present it "at an appropriate time". Currently, there is no public announcement regarding the evolution of GraalVM's AI contribution policy.

Original link:

https://www.infoq.com/news/2026/06/oracle-genai-policies/

This article is from the WeChat official account "InfoQ" (ID: infoqchina). Author: Karsten Silz, Editor: Tian Chengz. Republished by 36Kr with authorization.