CloudSEK’s report details a large-scale supply chain attack in which a compromised package exposed cloud keys, repository tokens, SSH keys, Kubernetes secrets and AI provider keys across thousands of downstream projects. Unit 42, meanwhile, has flagged a more AI-specific consequence of this trend: “token jacking”, where stolen AI access tokens are hijacked and routed to gray-market resellers who monetise the stolen compute.
Together, the two reports point to the same underlying problem — as companies grant broader AI access to speed up development, that access is becoming one of the most valuable and least protected assets in the modern software supply chain.
Cybersecurity player CloudSEK’s recent blog spoke about a hack that began in March 2026 but remains active. The threat actor group TeamPCP orchestrated what is believed to be the largest supply chain attack.
CloudSEK said the stolen objects included cloud keys, repository tokens, SSH keys, Kubernetes secrets, package-publishing credentials, environment variables and AI provider keys, which could allow attackers to move far beyond the affected package.
“A package can disappear in minutes while copied credentials remain usable for weeks or months unless they are rotated and downstream activity is investigated,” said the blog.
Importantly, the CloudSEK report stated that AI systems are becoming high-value junctions between data, identity, compute and autonomous action.
What is token jacking?
Just as companies were beginning to get a handle on the tokenmaxing issue — where AI usage costs spiral due to inefficient token consumption — they now find themselves dealing with an entirely new threat: “token jacking”, a process in which hackers hijack legitimate AI access tokens and route them to the gray market to monetise them.
While this is another form of hacking, the impact it creates can be significant for companies, according to Unit 42 by Palo Alto Networks. In a recent report, the company said it had found instances where companies lost millions of dollars in charges before the attacks were discovered and contained.
Cdr Raj Shastrakar (retd), director and head of Unit 42, India and SAARC, said AI token jacking is a new spin on a familiar attack: stealing access to computing resources. Attackers harvest long-lived credentials, such as application programming interface (API) keys for AI platforms, and use them to consume or resell that computing power through intermediary proxies known as “transfer stations”, often at a steep discount to legitimate access.
“What makes this particularly challenging in AI environments is the level of access organizations may grant in order to encourage experimentation and development. Developer accounts can carry high spending limits and broad permissions to create API keys, provision models or build new applications,” said Shastrakar.
He further added that this can have serious repercussions if a privileged account is compromised. “The potential impact therefore extends well beyond someone simply using a stolen token; the extent of the damage ultimately depends on the permissions attached to that account or credential.”
Unit 42 is seeing this threat grow rapidly as AI adoption creates an increasingly lucrative gray market for stolen compute access. At the same time, attackers are becoming more effective at scaling credential theft.
Why is software supply chain risk rising?
A recent Kaspersky study also found that supply chain attacks have become one of the most common cyber threats facing businesses worldwide, with nearly one in three organisations experiencing a supply chain-related incident over the past year.
“We see a rising volume of threats targeting open-source software. In 2025, we detected 19,484 malicious packages, a 37% increase from 14,197 in 2024, while hacktool detections rose 11% year-on-year from 2,966 to 3,302. The findings underscore the growing need for organisations to strengthen software supply chain security as open-source components become increasingly integral to modern applications,” said Sergey Lozhkin, head of APAC and META research units at Kaspersky GReAT.
With AI adoption accelerating and credential-based attacks growing more sophisticated, experts say organisations will need to tighten how they issue, monitor and rotate access tokens to stay ahead of this emerging threat.
 "cyber fraud")