Artificial intelligence is rapidly reshaping how enterprise software is built, but corporate security frameworks are struggling to keep pace with the change, according to a report released by software supply chain platform JFrog Ltd..
The JFrog 2026 Software Supply Chain Security State of the Union report indicates that while companies report high confidence in their cybersecurity measures, their actual risk exposure is expanding dramatically. This "illusion of mastery" stems from a widening gap between stated corporate policies and the operational reality of handling AI-generated code, open-source packages, and cloud credentials.
Tsunami of software data
Driven by AI-assisted development, artifact creation hit record levels by the end of 2025. The JFrog platform held 18.2 billion software artifacts at year-end, marking a 136% increase from 2024.
The explosion is largely fueled by AI and machine learning workloads. Hugging Face, a prominent AI model repository, published 1.4 million new packages in 2025 alone, making it the second-largest source of new software components tracked globally, trailing only Docker Hub.
Why fleet managers should care
While software code might seem far removed from moving freight, modern logistics relies heavily on connected data systems. Fleet management platforms, routing software, and automated dispatch applications are all built using these exact software pipelines.
"Every enterprise is adding AI to their software supply chain, which is increasing the attack surface for bad actors," said JFrog CEO & Co-Founder Shlomi Ben Haim. "Our report shows attackers are no longer just breaching traditional defenses – they are actively weaponizing the trusted models, registries, and agentic tools driving today's AI-powered development. The era of 'scan and hope' is over."
A compromised software supply chain means that the very tools trucking managers use to track vehicles, optimize fuel consumption, and monitor driver logs could become vectors for cyberattacks.
According to the report, the risk is compounding across multiple areas:
- Vulnerability spikes: More than 48,000 new software vulnerabilities were disclosed in 2025, a 20% increase over the previous year. Researchers attributed part of this surge to AI-generated code written by developers without applying standard secure coding practices, leading to decades-old security flaws like database injection.
- Malicious model payloads: The JFrog security team identified 495 malicious models on Hugging Face carrying live, hidden payloads. These packages included system command execution scripts and credential harvesting tools.
- Exposed cloud tokens: Researchers scanned public software repositories and found 17,637 exposed corporate security tokens and API keys. Crucially, 33% of the exposed Amazon Web Services (AWS) credentials found were confirmed to be still active, giving attackers direct potential access to corporate cloud infrastructure, data centers, and compute storage.
Governance gap
The survey of 1,508 security, development, and operations professionals highlighted significant corporate vulnerabilities. While 97% of organizations claim to have certified AI model governance, the prevalence of malicious models and unsecured entry points calls that figure into question.
Furthermore, 18% of surveyed organizations acknowledged they enforce no active governance over the internal developer tools and extensions sitting directly inside their engineers' coding environments.
“AI has not only changed how software is written; it has also increased the speed and scale at which zero-day vulnerabilities are exploited, and malicious software supply chain attacks are developed and distributed,” said Yoav Landman, JFrog CTO and Co-Founder. “To stay ahead, organizations need automated governance that curates every software asset entering the organization, whether introduced by agents or developers, and continuously monitors every release that contains those assets. The race is no longer about who discovers a zero-day first, because that information is advertised within minutes. It is about who can fortify their software supply chain at scale to keep their organization secure.”
Sourcing controls were built for traditional package registries, and for most organizations, that is still where they stop, the report notes, warning that corporate perimeters have expanded far faster than automated defensive tracking.
