Libreboot reverses AI contribution ban for risk-based review
The firmware project moved from an absolute ban on AI-assisted code to a policy of selective review to match the reality of upstream software dependencies.
The short version
- Maintainer Leah Rowe deleted the AGENTS.md file which previously banned AI-assisted contributions.
- The new Slop Reduction Policy replaces a total ban with a commitment to revert or rewrite low-quality AI patches.
- Libreboot admits it cannot purge all AI code because it relies on upstream projects like coreboot and EDK2.
- The project will now focus human review on high-risk areas like power management and security-sensitive drivers.
Why it matters
This shift highlights a growing reality for open-source projects: it is nearly impossible to maintain a 'pure' human-coded codebase when relying on external dependencies. For users and developers, it signals a move toward pragmatic risk management over ideological purity in software supply chains.
The shift from ban to review
Libreboot maintainer Leah Rowe officially removed the project's anti-AI contribution ban on November 20, 2026. This action followed a month of tension regarding the project's stance on using large language models (LLMs) to generate code. The removal was made explicit in a Codeberg commit where Rowe stated she would no longer enforce the ban.
The decision was driven by a realization that the previous policy was unenforceable. Rowe noted that she did not want to be a hypocrite by making promises she could not deliver as the industry moved toward AI-assisted development. By deleting the ban, she acknowledged that exceptions would be inevitable as upstream projects began shipping AI-assisted code.
While the ban is gone, the project did not embrace AI development without limits. Instead, it replaced the prohibition with a new document titled the AI Slop Reduction Policy. This new framework focuses on the quality of the output rather than the tools used to create it, aiming to filter out low-quality contributions while allowing for practical progress.
The reality of upstream dependencies
A primary reason for this policy reversal is the structural reality of how firmware projects are built. Libreboot is a downstream distributor that builds on top of projects like coreboot, EDK2, and various GNU projects. These upstream sources are already moving toward more relaxed norms regarding AI assistance, making a local ban difficult to maintain.
Rowe explained that banning AI locally while rebuilding upstream trees that already contain AI-touched code creates an unauditable inconsistency. If a project cannot fork every single dependency to remove AI elements, a total ban becomes a form of policy theater. This mirrors challenges faced by enterprise teams who may forbid internal AI tools while still importing third-party software trained on unknown data.
The project's documentation admits that it cannot purge all AI code while tracking these major dependencies. Because Libreboot does not author the entire stack from scratch, it must find a way to manage the code it inherits. The new policy acknowledges this limitation by focusing on maintainer judgment rather than automated rejection of specific tools.
Risk-tiered review for firmware
The new Slop Reduction Policy introduces a risk-based approach to code review. Instead of a blanket rule, the project will prioritize human attention based on the potential impact of a change. High-risk areas, such as power management, security-sensitive drivers, and hardware blobs, will face intense scrutiny or be reverted if they contain AI-generated fixes.
Conversely, the policy suggests a more tolerant stance toward low-risk paths like documentation and tooling. This tiering is designed to preserve scarce maintainer resources for the parts of the firmware that could actually brick a laptop. Since a bad firmware image can leave a machine unbootable, the price of an unreviewed patch is significantly higher than in a typical web application.
The policy even includes specific examples of how this will work in practice. For instance, when a specific driver included an AI-generated fix for Bluetooth, Libreboot chose to remove the entire driver rather than ship the AI-authored path. This demonstrates a preference for removing features non-destructively rather than accepting potentially buggy AI repairs.
Governance lessons for OSS
Libreboot's experience provides a blueprint for how other open-source projects might handle AI integration. It suggests that maintainers should focus on review obligations rather than making absolute oaths about authorship. By moving to a risk-tiered model, projects can maintain security without creating unenforceable rules that ignore the reality of modern development workflows.
The project's shift also highlights the importance of documenting partial resistance. Rather than silently accepting AI code or silently rejecting it, the new policy explicitly names examples of AI slop and the actions taken to mitigate it. This transparency helps contributors understand the project's standards while still allowing for the use of modern productivity tools.
For organizations looking at supply chain integrity, this move suggests a shift toward requesting build manifests and artifact reviews. Instead of relying on slogans about human-only code, stakeholders can demand evidence of human review on critical paths. This pragmatic approach ensures that the most dangerous parts of the software remain under human oversight while allowing the rest of the ecosystem to evolve.
What to watch
- Whether other downstream firmware distributions adopt similar risk-based review models.
- How the community reacts to the removal of the AGENTS.md ban.
- Whether Libreboot eventually forks any specific upstream projects to 'de-slop' them.
Try it
Check the Libreboot.org website for the 'AI Slop Reduction Policy' to see the specific examples of which drivers they chose to remove versus keep.
Sources and links
- Reportingexplainx.ai: the article we sourced
Read the original at explainx.ai
First reported by explainx.ai. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.