Lily: an automatic backdoor hunter for the commit and release stages

Hidden code-level backdoors — inconspicuous changes that unlock secret privileges via a covert «trigger» — remain a persistent threat to open-source software. Known attempts to sneak such code into widely used projects through malicious commits, tampered release packages or compromised third-party dependencies have, in essence, been stopped only by luck and manual review. The authors of a new paper presented at the ASE 2026 conference in Munich point out that ordinary continuous-integration pipelines fail to catch these attacks, while consumer-side binary-analysis tools demand serious manual effort.

Dimitri Kokkonis, Mickaël Marcozzi and Stefano Zacchiroli propose a tool called Lily, meant to harden development and release processes against backdoor injection. The detection mechanism plugs into two places at once: into CI pipelines, to block malicious commits, and into release-review procedures, so that tampered builds or infected dependencies don't slip into major ecosystems such as Linux distributions.

How it works

The paper's first contribution is an extension of CI-compatible fuzzing. Lily learns to recognize triggers of suspicious behavior by comparing historical and current runs of a program. According to the authors, this yields fast and accurate detection, suitable both for CI and for validating updates.

The second contribution is combining code-change analysis with fuzzing data. This makes it possible to point maintainers to the specific stretches of code that give away a backdoor, even when a release update touches millions of lines. In other words, the tool doesn't just say «something is wrong here» but also shows exactly where, saving hours of manual triage.

The authors don't confine themselves to the ideal scenario: they describe five strategies an attacker might use to try to evade Lily, and they assess the corresponding countermeasures.

Results

The experiments covered hundreds of both «clean» and infected commits and releases. According to the researchers, Lily achieves high detection accuracy with a low false-positive rate, reliably finds malicious code, resists evasion attempts — and, most tellingly, would have prevented real backdoor incidents that have occurred in the past. Given how much attention open-source supply-chain attacks have drawn lately, an attempt to automate what previously only the vigilance of individual maintainers could save looks timely.

Источник: Lobsters