Libreboot Drops Its Anti-AI Policy After One Month, Keeps Slop Reduction Libreboot maintainer Leah Rowe deleted the project's month-old anti-AI contribution ban from its lbmk repository on November 20, 2026, permitting AI-assisted submissions again, according to the Codeberg commit and Phoronix reporting. Rowe said she would not enforce the ban because upstream code and industry direction make exceptions inevitable, and Libreboot replaced it with an AI Slop Reduction Policy that reverts or rewrites AI-generated patches where feasible rather than promising a total ban. The policy cites a thinkpad_acpi case in which Libreboot removed the driver rather than ship an upstream AI-generated WWAN/Bluetooth fix, noting battery threshold automation regressed. Libreboot maintainer Leah Rowe deleted the project’s month-old anti-AI contribution ban on November 20, 2026 , according to the Codeberg commit https://codeberg.org/libreboot/lbmk/commit/cbbc12b86d?ref=explainx and Phoronix reporting https://www.phoronix.com/news/Libreboot-Drops-AI-Policy?ref=explainx . AI-assisted submissions are permitted again . In parallel, Libreboot published an AI Slop Reduction Policy https://libreboot.org/news/clankers.html?ref=explainx that explains resistance to low-quality AI patches without promising a total ban upstream projects already violate. For developers shipping open firmware or agent-generated drive-by PRs , the story is governance physics: downstream rebuilders inherit upstream norms. TL;DR | Question | Answer Nov 20, 2026 | |---|---| | What changed? | AGENTS.md AI ban deleted from lbmk | | Who decided? | Leah Rowe , Libreboot maintainer | | Why? | Policy called unenforceable vs coreboot and peers | | Replacement | AI Slop Reduction Policy on libreboot.org | | AI submissions now? | Allowed ; review standards softened but documented | | Upstream reality | coreboot and others do not reject AI contributions | Timeline: from AGENTS.md ban to deletion Pinned policy document beside a folder, evoking AGENTS.md contribution rules | Date | Event | |---|---| | Oct 2026 | Libreboot adds anti-AI policy in lbmk; language later refined in AGENTS.md | | Nov 20, 2026 | Leah Rowe commits delete the AI policy , removing AGENTS.md | | Nov 20, 2026 | libreboot.org publishes AI Slop Reduction Policy article | | Nov 20, 2026 | Phoronix and forums amplify the reversal | The speed of the reversal matters: Libreboot is not a Fortune 500 compliance department. It is a small downstream firmware distributor whose maintainer can rewrite policy when upstream reality contradicts a manifesto. The October ban and November reversal Phoronix recounts that Libreboot added an aggressive anti-AI policy in October 2026, initially with harsh language, later refined in AGENTS.md. On November 20 , Rowe removed the file entirely. Her commit message, as quoted by Phoronix, states she will not enforce the ban, expects to make exceptions as upstream ships AI-assisted code, and refuses to promise what she cannot deliver . The deletion is explicit: delete unreliable promise . That is a rare public admission that LLM contribution bans on small downstream projects collide with coreboot , EDK2 , and other dependencies Libreboot cannot fork at scale. What the Slop Reduction Policy actually promises Libreboot.org’s November 20 policy article titled around AI Slop Reduction replaces absolute prohibition with effort where feasible : - Revert or rewrite patches when AI-generated fixes slip through. - Sometimes remove features non-destructively rather than ship AI-authored repairs. - Acknowledge Libreboot cannot purge all AI code while tracking coreboot and GNU projects that already accept or tolerate LLM assistance. The article cites a thinkpad acpi example: upstream merged support with an AI-generated WWAN/Bluetooth fix; Libreboot removed thinkpad acpi rather than ship that path, while noting battery threshold automation regressed. That is materially different from a CI bot that rejects any @cursor commit: it is maintainer judgment on a tiny project surface. Gray blobs on a belt beside one green gem, evoking AI slop mixed with solid patches Rowe’s stated reasons primary source The November 20 Codeberg commit message, relayed by Phoronix, gives maintainer intent in her own words: - She will not enforce the ban going forward. - Upstream code and industry direction make exceptions inevitable. - She does not want to be a hypocrite selling an unenforceable promise. - She still believes the anti-AI essay content, but treats deletion as honesty. That distinction— personal opposition versus project enforceability —is the heart of the story. Libreboot did not publish a pro-LLM manifesto; it admitted a governance mistake and replaced it with a narrower, feasible policy. Why upstream matters more than manifestos Libreboot’s policy page states plainly that coreboot does not reject AI-generated contributions , and that GNU and EDK2 ecosystems are moving toward lax review norms. Libreboot is a downstream coreboot distribution ; banning AI locally while rebuilding AI-touched upstream trees creates unauditable inconsistency . The same structural problem appears in enterprise monorepos when one team forbids Copilot but imports vendor SDKs trained on unknown corpora—policy theater without supply-chain teeth. Libreboot versus LibreOffice-style mega-forks Rowe’s article contemplates forking projects to de-slop them, then admits Libreboot cannot fork every upstream. That is the same calculus small Linux distributions face when maintainers inherit AI-touched churn they did not author. A total AI ban only works when you own the compiler, libc, and VCS history . Libreboot builds coreboot + payloads ; it does not author the entire stack. Enterprise parallel: “no Copilot” teams importing SaaS Corporations clone Libreboot’s failed pattern when security policy says no GitHub Copilot but procurement buys SaaS with opaque training data . The fix is not a stronger PDF; it is artifact review on paths you ship. Libreboot’s Slop Reduction Policy is closer to risk-tiered review : - High risk: power management, blobs, security-sensitive drivers → scrutinize or revert. - Low risk: docs tooling → tolerate more automation. That tiering is a familiar pattern: the more a change can affect boot integrity, power behavior or hardware safety, the more a human reviewer should read it line by line. Low-risk paths such as documentation tooling can carry a lighter touch. The point is to spend scarce maintainer attention where a mistake would be hardest to detect and most costly to users who flash the result onto real machines. For a firmware project this matters more than for a typical web app. A bad firmware image can leave a laptop unable to boot, and recovery often needs an external flasher. That raises the price of every unreviewed patch, whoever or whatever wrote it. It also explains why Rowe chose selective reverts and removals, such as dropping thinkpad acpi, over a blanket promise she could not keep. Lessons for open-source AI governance 1. Ban only what you compile yourself — If you do not vendor the dependency, you cannot attest its authorship chain. 2. Prefer review obligations over oaths — Rowe’s reversal highlights hypocrisy risk when AGENTS.md promises total bans. 3. Document partial resistance — Slop reduction with named examples beats silent acceptance or silent rejection. 4. Expect public scrutiny — Phoronix and firmware forums will debate whether a downstream ban matched upstream reality; your contributors will too. Teams debating Libreboot-style bans https://explainx.ai/blog/haproxy-willy-tarreau-ai-bug-reports-maintainers-opus-glm-mythos-oss-scanner-cla-2026 in other infra projects should read Rowe’s commit as data , not dunk fodder. Comparison to other OSS AI policies | Project stance | Enforcement surface | Libreboot lesson | |---|---|---| | Total ban old Libreboot AGENTS.md | lbmk only | Collapsed when upstream ignored it | | Slop reduction current | Revert/rewrite selective patches | Honest about limits | | coreboot upstream | Accepts AI contributions per Libreboot | Sets ceiling for downstream | | Corporate CLA + AI scanner | Legal + CI | Works only with owned code | Will Libreboot scan future commits for AI? The Slop Reduction Policy does not promise automated detection. It promises maintainer effort where feasible —reverts, selective removals, occasional forks. There is no CI step described on libreboot.org that blocks commits based on LLM detectors. If you need machine assistance, combine: - Signed-off-by discipline Rowe uses signed commits on the deletion . - Reproducible build hashes comparing downstream blobs to upstream releases. - Human review on drivers that brick laptops when wrong—Libreboot’s audience still runs bare metal. AGPL ecosystems and agent skills Libreboot itself is not an AI product, but readers comparing AGPL agent chat https://explainx.ai/blog/agentes-chat-open-source-self-hosted-slack-alternative-ai-agents-mcp-agpl-github-trending-2026 stacks should note: license choice does not solve authorship attribution . AGPL forces source sharing; it does not prove whether a patch was LLM-drafted. What maintainers can do Monday - Require project files or reproducible build artifacts for sensitive drivers when AI authorship is suspected—publishers facing AI cover art disputes increasingly ask artists for layer files for the same reason. - Label AI-assisted PRs in CONTRIBUTING.md without pretending git history is pure. - Invest in human review for power management and security paths where bad AI patches brick hardware. Related reading Sources Primary: Libreboot lbmk commit cbbc12b86d https://codeberg.org/libreboot/lbmk/commit/cbbc12b86d?ref=explainx November 20, 2026 . Policy text: AI Slop Reduction Policy https://libreboot.org/news/clankers.html?ref=explainx Libreboot.org, November 20, 2026 . News coverage: Phoronix, November 20, 2026 https://www.phoronix.com/news/Libreboot-Drops-AI-Policy?ref=explainx . FAQs in depth: common reader questions Does deletion mean Libreboot endorses LLM coding tools? No primary source says that. The commit deletes enforcement machinery while the Slop Reduction article restates ideological opposition. Practically, contributors may use AI; maintainers may revert sloppy output. Should downstream distros republish the old AGENTS.md? Forks can, but they inherit the same upstream constraint. Banning locally without auditing coreboot imports creates a false sense of cleanliness . How does this interact with Secure Boot and reproducible builds? Libreboot’s value is freedom-respecting firmware . AI policy does not change measured boot properties; it changes trust in patch authorship during review. What about government or school laptop programs? Institutions buying Libreboot machines care about supply-chain integrity . They should ask vendors for build manifests , not slogans—especially after this policy swing. Will coreboot follow Libreboot’s old ban? Nothing in Rowe’s commit requires coreboot to change. Upstream governance is independent; Libreboot’s lesson is for downstream packagers , not a prediction about coreboot’s next AGENTS.md. Track coreboot mailing list threads if you ship forks that must certify patch provenance for auditors or procurement officers. Related policy posts on explainx.ai - ECHR and AI hallucinated citations for lawyers https://explainx.ai/blog/echr-romanian-lawyer-barred-two-years-ai-hallucinated-citations-november-2026 - Google workers petition on schools and AI search https://explainx.ai/blog/google-workers-petition-schools-search-without-ai-gemini-november-2026 - Signal screenshot blocking vs AI agents https://explainx.ai/blog/signal-blocks-screenshots-by-default-ai-agents-siri-gemini-recall-android-ios-macos-2026 Policy and repository state are accurate as of November 21, 2026.