# GhostJacking Exposes AI Agent Identity Gaps

> GhostJacking targets the gap between human and AI agent identity lifecycle management, enabling attackers to hijack orphaned agent sessions that retain access but are no longer monitored.

**URL:** https://www.ciptadusa.com/blog/ghostjacking-identity-governance-ai-agents-20260811  
**Type:** blog  
**Author:** PT Cipta Dua Saudara  
**Category:** Application Security  
**Published:** 2026-08-11  
**Cover:** https://cdn-uagents.enitip.com/uploads/blog/2026-08/daily-appsec-20260811-014720.jpg  

## Article

A new attack vector called **GhostJacking** exploits a fundamental gap in how organizations manage AI agent identities — not by cracking credentials, but by hijacking sessions that should be dead yet remain active behind the scenes.

## Summary

GhostJacking targets the gap between human identity lifecycle management and AI agent identity management, allowing attackers to hijack orphaned agent sessions that still hold access but are no longer monitored by anyone.

## The Challenge

When an organization deploys an AI agent — whether for DevOps automation, customer service, or security monitoring — that agent receives identity credentials: API keys, service accounts, OAuth tokens. The problem begins when an agent is decommissioned or reassigned.

**Traditional identity governance was designed for humans.** Offboarding an employee triggers a checklist: revoke the badge, disable the AD account, rotate shared credentials. But who "offboards" an AI agent? In most organizations, the answer is: nobody is responsible.

The result is an **orphaned identity** — a credential that's still active, still privileged, but no longer tied to a monitored process. GhostJacking exploits this window. Attackers don't need to steal new credentials; they just need to find old ones that were never revoked.

## Approach

**Three defense layers organizations need to build:**

**Identity lifecycle binding.** Every AI agent must have an explicit owner and expiration date on its credentials. Not a generic "rotate every 90 days" — but conditional expiry tied to the agent's deployment status. Agent removed from the orchestrator? Credential dies in 60 seconds, not 90 days.

**Behavioral baseline per agent identity.** A decommissioned agent should have zero activity. The first alert isn't "anomalous behavior" — it's "any behavior at all." If a service account tied to a dead agent suddenly makes an API call, that's not an anomaly. That's an indicator of compromise.

**Governance parity between human and non-human identities.** A CISO with an offboarding dashboard for 500 employees but no equivalent for 2,000 AI agent credentials is building a blind spot that widens every month.

This isn't a future problem. Organizations already running multi-agent systems today — and their numbers are growing rapidly as LLM-based automation adoption accelerates — need to audit their identity governance now, before GhostJacking becomes an actively exploited technique in the wild.

## References

- ['GhostJacking' Exposes Identity Governance Gaps in AI Agents](https://www.darkreading.com/cyber-risk/ghostjacking-identity-governance-gaps-ai-agents) — Dark Reading
- [NIST SP 800-207A: Zero Trust Architecture for Non-Human Identities](https://csrc.nist.gov/publications/detail/sp/800-207a/final) — NIST

---

*Markdown version of https://www.ciptadusa.com/blog/ghostjacking-identity-governance-ai-agents-20260811 — generated for AI agents and LLM crawlers.*