AI coding tools increase dependency sprawl, raising malware risks for developers

AI coding assistants are speeding up the growth of software dependencies and widening the attack surface for malware, a new analysis by security firm Chainguard found. The report, released Tuesday, says developers using tools such as GitHub Copilot, Tabnine and other large‑language‑model based generators are adding twice as many third‑party libraries to projects as they did a year ago. The surge in automatically suggested code is creating a sprawling “dependency sprawl” that security teams struggle to track.

The trend reflects a broader shift in software development. Since 2022, AI pair programmers have become mainstream, promising faster code writing and fewer bugs. At the same time, open‑source libraries have multiplied, and many contain known vulnerabilities. Supply‑chain attacks like the 2021 SolarWinds breach have made firms more wary of hidden risks, but the convenience of AI suggestions often outweighs caution. Developers accept generated snippets without fully vetting the underlying packages, leading to a cascade of unverified components in production code.

Chainguard’s data shows that 38 percent of new dependencies introduced by AI‑generated code have at least one known security flaw, and 12 percent are linked to previously identified malware. In one case, a popular AI tool suggested a snippet that pulled in a library with a backdoor, allowing remote code execution on vulnerable servers. Security researchers say the problem is compounded by the “black‑box” nature of the models, which can surface code from compromised repositories without warning. Industry groups are calling for stronger software bill of materials (SBOM) practices and for AI providers to audit the code they suggest.

The report urges developers to treat AI‑generated code as untrusted input, subject to the same review as any third‑party library. Chainguard recommends integrating automated dependency scanning into CI pipelines and using provenance data to verify the origin of suggested packages. Some AI tool vendors have pledged to add vulnerability checks before offering code, but experts warn that regulatory guidance may be needed to enforce consistent standards. As AI coding assistants become more entrenched, the balance between productivity and security will shape the next wave of software supply‑chain defenses.