The second release candidate of systemd 262 arrives one week after the first and, among fixes and refinements, introduces a detail that carries more weight than usual: an AI canary designed to surface code contributions generated by LLMs and not reviewed by a person before the pull request.

The concept, in its most common form, is straightforward: a visible directive is placed in project files or guidelines, and LLM-based coding assistants are instructed to preserve it. If a contribution arrives with the canary altered or missing, a signal goes off: that code may not have passed through human review.

systemd's move shifts the issue from peripheral libraries to core infrastructure. An init system and service manager is the first process to start and the last to stop on many Linux machines. An unreviewed contribution in such a component is not a style problem: it is a potential vector for regressions or vulnerabilities that spreads to every distribution and every environment that adopts it.

The canary is not a technical barrier. Its effectiveness depends on the model following the instruction not to remove it. Its real strength is different: it creates a signaling cost for anyone generating code with AI without disclosing it, and shifts the burden of proof toward the contributor. For maintainers, it reduces the need to hunt for generated patterns in every diff; for contributors, it makes explicit the line between human assistance and blind delegation.

For those managing on-premise or self-hosted infrastructure, this is a matter of audit and security posture. systemd is not an isolated application component: it is part of the layer that starts services, mounts filesystems and manages processes. The provenance of code landing at that level is not abstract, especially in regulated environments where supply chain documentation matters as much as functional tests.

Not everyone comes out ahead. Smaller projects with fewer review resources may adopt similar canaries without the capacity to respond to the alarms they generate. Contributors who use AI assistants opaquely risk having patches rejected that previously passed without questions.

The structural signal is that trust in open source code can no longer assume the absence of AI: it must be made verifiable, even with lightweight tools, before heavier controls become necessary.

systemd is not banning the use of LLMs, but making it more expensive to use them without oversight. It is a lightweight precedent, not a cryptographic attestation system: a low-cost signal to bring human review back to the center of the open source conversation.