Why OpenJDK Bans All AI‑Generated Code: Policy Details and Industry Contrast

OpenJDK Interim Policy on Generative AI

In April 2026 OpenJDK released a policy that forbids any contribution containing content generated by large language models, diffusion models, or similar deep‑learning systems, in whole or in part.

OpenJDK community contributions must not contain content generated by large language models, diffusion models or similar deep‑learning systems, in whole or in part.

The policy applies to source code, documentation, images, Git repositories, pull requests, mailing‑list posts, wiki pages, and JBS issues.

FAQ answer to the question “If I generate 100 lines with AI and edit 10, can I submit?” is “No”. Any AI‑generated content, regardless of proportion, is prohibited.

Private use of generative AI for understanding, debugging, reviewing existing code, or research related to OpenJDK is permitted.

Reasons for the ban

Reviewer burden

AI can produce code that passes tests but may contain design flaws or outright errors. Maintainers have limited time; a surge of AI‑assisted pull requests forces additional filtering.

New Relic 2026 report shows AI‑generated code often receives higher review scores yet leads to higher production‑environment failure rates, described as “looks good, breaks things”.

JDK is not a testbed

The JDK underpins critical systems in banking, telecom, government, and healthcare. The policy states:

Safety and reliability are paramount. Code that looks reasonable but is actually wrong directly threatens these attributes.

Intellectual‑property concerns

Contributors must sign the Oracle Contributor Agreement (OCA), guaranteeing ownership of each contribution. Most generative‑AI models are trained on copyrighted material, and the legal status of AI‑generated output is an unresolved litigation issue.

For Oracle, maintaining a clear, defensible provenance for every line of OpenJDK code is essential.

Oracle’s internal AI use vs. OpenJDK policy

Oracle publicly promotes AI‑assisted development internally while the OpenJDK policy bans AI‑generated contributions, creating a noted contrast.

GraalVM policy

GraalVM allows AI‑assisted contributions provided the submitter can defend every line of code. The core condition is:

The person submitting the change must be responsible for the entire contribution and be able to defend every line of code.

This mirrors the Linux kernel approach, which also permits AI assistance but requires full human responsibility.

Policies of other open‑source projects

OpenJDK

: complete ban on AI‑generated content QEMU: similar ban GCC: prohibits LLM‑written contributions Gentoo: prohibits AI‑assisted contributions NetBSD: treats AI‑generated code as copyright violation SDL: prohibits LLM‑generated code contributions Linux kernel: allows AI assistance but requires full responsibility GraalVM: allows AI assistance with full responsibility Apache: allows with conditions Debian: discussion ongoing, no decision yet

Conclusion

The OpenJDK ban functions as a pre‑marriage contract defining risk boundaries for AI tools before integration into a critical open‑source platform.

GraalVM’s statement—if a contributor cannot explain, defend, or maintain an AI‑assisted change, the change should be rejected—captures the core issue more directly than any blanket prohibition.