Express Computer
Home  »  News  »  Your dependencies are older, deeper and riskier than you think, and AI coding assistants aren’t helping

Your dependencies are older, deeper and riskier than you think, and AI coding assistants aren’t helping

0 2

For CIOs, the software supply chain problem is no longer limited to what developers knowingly install. A new report from CleanStart, a provider of verified software artifacts for modern software supply chains, suggests the larger exposure sits in what arrives indirectly, and in the AI tools now recommending packages to engineering teams.

CleanStart’s State of Software Dependency Trust 2026: From Dependency Management to Verified Software Foundations examines how dependency complexity, aging software, upstream attacks and AI-assisted development are changing the risks organisations face when building and maintaining software.

India’s developer boom raises the stakes

The findings land as India’s developer base expands quickly. According to the report, India added 5.2 million developers during 2025, reaching 21.9 million on GitHub. It is projected to reach 57.5 million by 2030, ahead of the United States at 54.7 million.

More developers shipping more software, increasingly with AI assistance, means more components entering enterprise environments. That raises the cost of weak controls over what gets adopted.

Open source is everywhere, and mostly indirect

Across audited commercial codebases, open source appeared in 98% of applications, with an average of 1,180 components per application. Of those:

64% entered indirectly through other dependencies, so teams never chose them directly.
68% of observed components were more than two years old.

A separate analysis of 947 audited commercial codebases found that 17% of observed components sat outside conventional package managers, and 41% were ten or more versions behind. The report also highlights how maintenance responsibility for open-source projects is concentrated among small developer teams.

For a CIO, those figures point to an inventory problem. Components that sit outside package managers, or arrive several layers down the dependency tree, are harder to track, patch and attribute.

AI recommendations are a weak link

The report’s most striking finding concerns AI-assisted development. Only 20% of tested AI-generated dependency recommendations were both valid and free of known vulnerabilities.

The report also points to a fix. Grounding AI models in live registry data reduced high- and critical-risk recommendations by roughly 70% in the tested scenarios. That suggests the issue is less about whether to use AI coding tools than about what data they are connected to when they suggest packages.

The vulnerability numbers need careful reading

The average number of vulnerabilities per audited codebase rose 107%, from 280 to 581. The share of codebases containing vulnerabilities moved only from 86% to 87%.

CleanStart cautions that this distinction matters: rising average vulnerability counts do not necessarily mean a comparable increase in affected applications. Nearly every codebase was already affected, and each now carries more findings. That makes prioritization, rather than raw counts, the practical challenge for security teams.

The fixes already exist

Approximately 95% of vulnerable component downloads examined in 2024 registry telemetry had a fixed version available at the time of download. In most cases the patch was available, but teams did not adopt it.

That points to a process and tooling gap rather than a shortage of fixes. The report also examines upstream attacks, including compromised publishing credentials, malicious install scripts and manipulated CI/CD infrastructure, as well as gaps in SBOM adoption and consumption.

From scanning to evidence

CleanStart argues that organisations should move beyond vulnerability scanning. Vijendra Katiyar, co-founder of CleanStart, said, “India has an opportunity to shape how the next generation of software is built, adopted and secured. As development accelerates and AI becomes part of everyday engineering workflows, organisations need to look beyond vulnerability scanning and understand the software components they are bringing into their environments. The focus needs to shift towards establishing verifiable evidence of software identity, origin, provenance and integrity before components are adopted. This is how organisations can build confidence into the software supply chain rather than relying only on finding and fixing problems later.”

The report recommends establishing minimum evidence requirements for widely reused or consequential software foundations, covering identity, origin, provenance, integrity, security, maintenance and policy. It adds that making this evidence reusable across developer and AI workflows can reduce repeated assessment efforts.

Regulation is moving the same way

India’s regulatory direction supports the shift. CERT-In’s SBOM v2.0 technical guidelines and explicit SBOM requirements within SEBI’s cybersecurity framework for regulated financial entities reflect the growing importance of software component transparency and accountability. For CIOs in regulated sectors, producing and consuming SBOMs is becoming an expectation rather than an option.

Leave A Reply

Your email address will not be published.