DayBlink Consulting Partner and Cyber Security Practice Lead Michael Morgenstern authored “The Growing Challenge of AI Agent & NHI Management”, examining the security risks posed by the rapid growth of AI agents and non-human identities (NHIs).
As AI agents rise in popularity and machine identities now outnumber human identities 82:1, organizations face an unprecedented and rapidly closing window to secure their agentic AI infrastructure. Drawing on live attack demonstrations at security conferences, this article argues that — for the first time in a long time — we have advance warning of probable attacks before they arrive. Agents operate with autonomy and broad access, making them uniquely dangerous if compromised, and the number of NHIs required to support them is growing exponentially.
Key Takeaways:
- AI agents rely on non-human identities that are poorly managed, often over-permissioned, and increasingly targeted — including through jailbreaking, prompt injection, data poisoning, and malicious proxy attacks already demonstrated in the wild
- Abandoned and orphaned agents — forgotten but still live and exposed — represent a vastly underestimated attack surface that organizations need protocols to identify, quarantine, and shut down
- Because LLMs are unpredictable, the NHIs that agents leverage can cause permissions to sprawl beyond direct organizational control, compounding the challenge of least-privilege enforcement
- A defense-in-depth approach is essential: fingerprinting agents with unique verifiable identities, filtering for prompt injection, sandboxing with least-privilege configurations, and using DLP and audit logging are among the ten concrete mitigation steps outlined
Read the full article here: Link