How Flyingduck's SBOM Stops Risk Before Merge

09 Jun 2026

|

10:00 UTC

Your SBOM told you that you were exposed. Within hours of the LiteLLM compromise in March 2026, organisations with well-maintained SBOMs could confirm the poisoned PyPI package was in their dependency tree. They knew they were affected. They knew which applications carried the risk.

sbom-cbom-aibom

What they could not answer was when the compromised package had entered their codebase, which PR had introduced it, or how to fix it without pulling a developer off current work, and cycling through another review process weeks after the vulnerable commit had already merged.

The TeamPCP threat actor group embedded credential-harvesting code in LiteLLM through poisoned packages and malicious pull requests. They harvested AWS keys and GitHub tokens, then monetised the access through ransomware partnerships. The SBOM did its job. It delivered visibility. But visibility alone did not prevent exposure from reaching production.

That gap, between seeing the risk and stopping it at the point of entry, is where most supply chain security programs stall.

The Visibility Stack: SBOM, CBOM, and AIBOM

Flyingduck's SBOM is built to integrate with your organisational stack. It maps software components, open-source packages, libraries, and third-party dependencies across your applications in real time. When a CVE drops, you know your exposure immediately. When CERT-IN's six-hour reporting window starts, you already have the answer.

But software components are only one layer of your technology risk surface. Two other inventories complete the full picture.

The CBOM (Cryptographic Bill of Materials) maps your encryption algorithms, digital certificates, cryptographic keys, and security procedures. As your organisation prepares post-quantum migration, CBOM tells you exactly where legacy cryptography lives, which systems depend on it, and what needs to move first. Without it, the transition to post-quantum standards is a guessing game with compliance deadlines attached.

AIBOM (AI Bill of Materials), on the other hand, tracks the AI models, training data, software frameworks, and third-party AI services running across your stack. As regulators from the EU AI Act to India's emerging AI governance framework start calling for transparency on how AI systems are trained, deployed, and monitored, AIBOM gives you the audit trail before the auditor asks for it.

Together, SBOM, CBOM, and AIBOM give your organisation a unified visibility across the three most dynamic and difficult-to-track technology layers: software supply chain, cryptographic infrastructure, and AI systems. This perceptibility is the foundation. No serious security program operates without it.

The question is what you do with it.

Visibility Is the Foundation. Interception Is the Control.

Here is the scenario that keeps this conversation honest. Your organisation maintains SBOMs in line with NIST guidance and Executive Order 14028. Your CBOM is current. Your AIBOM tracks every model in production. You meet CERT-In's six-hour reporting mandate because you can identify affected components fast. Your next audit goes clean.

And a vulnerable transitive dependency still entered your codebase six sprints ago through a PR that nobody flagged.

Visibility tells you about the risk that exists. It does not tell you when the risk was entered, which commit introduced it, or how to fix it within the workflow, while the developer is still in context. By the time a build-stage scan or a post-release audit surfaces the issue, remediation means a new ticket, a context switch, and a fresh PR cycle for a problem that could have been caught at the commit where it was introduced.

Flyingduck pairs your visibility stack with commit-stage detection. The platform scans code at the commit, surfaces the specific line and dependency path that introduced the risk, and provides code-level remediation guidance inside the same PR. A differential scan verifies the fix before merge. The developer never leaves their workflow. The vulnerable code never reaches the main branch.

Three Layers Deep: The Transitive Dependency Problem

Modern applications carry hundreds of transitive dependencies, packages your code never imports directly but inherits through the libraries it does use. The Synopsys 2024 OSSRA report found that 96% of commercial codebases contain open-source components. What it did not quantify is how many of those components arrived as transitive imports that no developer explicitly chose or reviewed.

Flyingduck's SBOM maps the full dependency graph, including transitive chains, in every PR. But mapping alone is not enough. An attacker scanning the same tree, whether manually or with AI-assisted tooling, does not care whether a vulnerable package sits in your direct dependency list or four layers deep. They scan for exploitability. Your visibility layer needs to do the same.

This is where RBVM (risk-based vulnerability management) separates Flyingduck from SBOM-only approaches. The platform determines reachability, assesses whether a flagged CVE is exploitable in your specific code context, and prioritises accordingly. Your developers fix what matters, not everything that appears on a flat list. SBOM tells you what you shipped. RBVM tells you what is exploitable, what is reachable, and what to fix first.

From Visibility to Interception

Your organisation needs a visibility stack. SBOM for software components. CBOM for cryptographic assets. AIBOM for AI systems. Flyingduck delivers all three, integrated into your organisational workflow rather than bolted on as an afterthought.

But visibility is the starting line. Flyingduck's commit-stage detection, transitive dependency analysis, PR-stage provenance context, and pre-merge verification turn that visibility into an active control. The bill of materials is no longer a post-build artefact you hand to an auditor. It becomes a living layer inside a pipeline that catches risk where it enters: the commit.

Security teams that pair visibility with commit-stage interception own the outcome. Teams that stop at the inventory layer can see the problem. They cannot stop it before the merge.