The industry-wide report examines how legacy components, upstream attacks, and AI-assisted development expose gaps in software supply chain security.
Key points:
- AI dependency risk: Among the AI-generated dependency recommendations tested, only 20% were both valid and free of known vulnerabilities.
- Dependency complexity and age: The audited applications contained an average of 1,180 open-source components, 64% of which were introduced indirectly through other dependencies.
- Opportunity for remediation: Of the vulnerable components downloaded in the 2024 registry telemetry, approximately 95% had available patches at the time of download.
San Jose, California, November 17 /PRNewswire/ -- CleanStart, which provides verified software artifacts for modern software supply chains, today released "The State of Software Dependency Trust 2026: From Dependency Management to Verified Software Foundations." The report examines how rising dependency complexity, component aging, upstream software supply chain attacks, and AI-assisted development are reshaping the risks organizations face when building and maintaining software.
The report also found that 68% of components in the audit sample had been in use for more than two years; another registry telemetry analysis revealed that approximately 95% of vulnerable components had available patched versions at the time of download. Taken together, these findings indicate that the challenges facing software teams are growing: they must not only understand what is entering their applications but also assess its trustworthiness, while continuously maintaining dependencies as developers and AI systems introduce them at an increasingly rapid pace.
Nilesh Jain, CEO and co-founder of CleanStart, said: "Software security cannot stop at merely identifying vulnerabilities. Organizations need to know what they are bringing into their environment, where it comes from, and whether its integrity can be verified. As software development accelerates, especially driven by AI, the volume and speed of dependencies being adopted will only continue to grow. The industry needs to shift from managing known risks after adoption to building confidence in software before use."
In this ten-part analysis, the report examines software composition, patch delays, industry and regional differences, upstream attacks, maintainer capacity, and AI-assisted development. Other findings include:
- Of the 947 audited commercial codebases, 17% of observed components were outside traditional package managers, and 41% were 10 or more versions behind. Another review of 47 leading non-npm open-source projects found that in 40% of projects, a small number of developers contributed the majority of commits, highlighting broader visibility and maintenance challenges.
- The average number of vulnerabilities per audited codebase increased by 107%, rising from 280 to 581. However, the percentage of codebases containing vulnerabilities increased only from 86% to 87%, indicating that the rise in average findings does not correspond to a proportional increase in the number of affected applications.
- Combining AI models with real-time registry data reduced high-risk and severe-risk recommendations by approximately 70%. The report also explores how security tools enhance the quality of AI-generated dependency recommendations.
- The report examines software supply chain incidents involving compromised release credentials, maliciously installed scripts, and manipulated CI/CD infrastructure. It also highlights the scale of malicious package activity and the challenges organizations face in generating, using, and enforcing SBOM requirements.
The report recommends starting with the most widely reused or most critical images, libraries, runtimes, and packages. Organizations should establish minimum evidence requirements for these foundational components, covering identity, origin, provenance, integrity, security, maintenance, and policy. Enabling this evidence to be reused across applications and accessible through existing developer and AI workflows can reduce redundant evaluation efforts by individual teams.
CleanStart implements this model through Clean Images and Clean Libraries, providing verified, reusable software foundations. This approach complements continuous scanning, SBOM generation, remediation, and runtime controls, and is continuously reviewed as components, threats, and organizational requirements evolve.
The report synthesizes published audits, investigations, ecosystem telemetry, incident reports, and policy sources, with full references and explanations of methodological limitations.
The full report is available at https://clnstrt.co/dependency-trust-2026.
About CleanStart
CleanStart helps organizations secure modern software supply chains with verified software artifacts. Companies enable organizations to discover software assets, establish software provenance, replace insecure components with verified alternatives, and continuously verify software integrity. This empowers organizations to build, procure, and deploy software with greater confidence while strengthening their software supply chain security posture.
CleanStart was founded by Nilesh Jain, Vijendra Katiyar, and Biswajit De, all of whom bring over 20 years of global leadership experience in cybersecurity. CleanStart is working to make “verification” the foundation for building and securing modern software.
Media Contact:
Kyle Porter
EVP-Managing Director
[email protected]

