← BACK TO FEED
OracleOpenJDKAI code generationLarry Ellisonopen source

Oracle Bans AI Code From OpenJDK While Larry Ellison Brags His Entire Company Runs On It

Oracle has banned AI-generated code and content from contributions to the open-source OpenJDK project, citing concerns about reviewer workload, safety, security, and intellectual property risks, while still permitting developers to use AI tools privately for tasks like debugging and research. This stance stands in notable contrast to Oracle's internal enthusiasm for AI-generated code, with co-founder Larry Ellison boasting that AI now writes Oracle's own software, and the company attributing recent mass layoffs partly to AI adoption. The apparent double standard has drawn scrutiny, coinciding with Oracle's aggressive $70 billion datacenter investment and a subsequent credit rating downgrade to just one notch above junk status.

Oracle has quietly told OpenJDK contributors to keep AI-generated content out of the project entirely. That means no LLM-produced source code, no AI-written documentation, no generated anything in pull requests, bug reports, wikis, or email threads. The ban covers the lot.

The company framed it as an interim measure while a fuller policy gets drafted, citing reviewer burden, security risks, and the usual intellectual property fog that hangs over anything touching generative AI. The underlying concern is straightforward enough: OpenJDK underpins mission-critical infrastructure across a huge chunk of the industry, and plausible-looking but subtly wrong code is arguably more dangerous than obviously broken code.

Developers can still use AI tools privately, for reading code, debugging, or research. They just can't feed the output back into the project.

All reasonable so far. Then you look at what Oracle's own executives have been saying.

At Oracle AI World 2025, founder and CTO Larry Ellison announced, with characteristic modesty, that Oracle doesn't actually write its own software anymore. "The code that Oracle is writing, Oracle isn't writing," he said. "Our AI models are writing." Co-CEO Mike Sicilia echoed the enthusiasm earlier this year, describing AI coding tools as central to Oracle's ability to ship faster with smaller teams.

Oh, and Oracle cut 21,000 jobs in June, citing AI deployment across its operations as a contributing factor. So the technology is apparently reliable enough to replace thousands of engineers, but not trustworthy enough to appear in an open source Java repository.

The gap between those two positions is wide enough to drive a datacenter through.

Oracle hasn't explained what makes its internal AI-generated code categorically safer than anything a contributor might submit to OpenJDK. Perhaps it has some internal review process that magically eliminates the safety and security risks it flagged in its own policy statement. Perhaps the standards simply differ when the code is yours versus someone else's. Neither answer is particularly satisfying.

The timing adds another layer of awkwardness. Oracle pledged $70 billion in datacenter spending over the coming year, up from $55.7 billion in fiscal 2026. S&P responded by downgrading Oracle's credit rating to BBB-, one notch above junk, with a sceptical note about whether the returns would materialise. Publicly casting doubt on AI reliability right now would not be ideal for a company betting its balance sheet on the technology delivering.

So the ban on AI contributions to OpenJDK may be entirely sensible on its own terms. The problem is that Oracle's broader messaging makes it very hard to take the stated reasoning at face value.

READ NEXT
The Anti-AI Open Source Movement Has One Thing Going For It, and Everything Else Working Against ItCapital One Releases AI Vulnerability Hunter to the PublicLinus Torvalds on AI Coding Critics: Fork Off