The Non-Human Identity Blind Spot: What Cisco's Astrix Acquisition Reveals About Your Agentic Attack Surface

Here is how a typical NHI compromise unfolds. An attacker finds a hardcoded API key in a public GitHub repository. That key belongs to a service account with access to a cloud storage bucket. In that bucket is a Terraform state file containing the credentials for a Kubernetes cluster. Once inside the cluster, the attacker discovers a ConfigMap with a database connection string. The database contains a table with OAuth refresh tokens for the company's internal CRM system. The attacker uses those tokens to exfiltrate customer records. Each step is a legitimate credential being used for its intended purpose. No alerts fire because the activity looks normal. The only way to catch this is to understand the dependency graph between identities and resources. The exploit path exists because of the relationships, not any single misconfiguration. You can write a Bash script that enumerates these relationships by scanning your cloud provider's IAM policy simulator. Here is a Python function that does the basics: it takes an AWS role ARN, simulates what it can access, and recursively follows each service it can reach. Running this against your production environment will probably surprise you. The code is trivial to write but most organizations never run it because they assume their vault handles everything.

Start with discovery. The identity team should run a script that queries your cloud provider for all service accounts, their creation dates, last authentication times, and permission boundaries. Do this for every provider you use. Most teams skip this because they think their CMDB covers it. It does not. The CMDB only knows about active applications. Dead integrations leave ghosts. Second, the security engineering team should implement a credential expiration policy that applies to all non-human identities. No exceptions. Service account keys must expire within 90 days. OAuth tokens must have a maximum lifetime of 30 days. Third, the platform team must integrate secret scanning into the CI/CD pipeline as a blocking gate. Not just for public repositories. For private ones too. Attackers who breach your internal network will scan your internal repos. Fourth, once per quarter, the IAM team should run a permission review for every service account with access to production. This is the step most teams skip. They review human access quarterly but rarely review service accounts. Fifth, build a dependency map of your credential relationships. This is the most important step and the one nobody does. Know exactly which service depends on which credential and what that credential unlocks. You can use a tool like a custom Python script that parses your deployment manifests and IAM policies to generate this graph. When an attacker compromises one credential, you need to know the blast radius. If you cannot answer that question in under five minutes, you have a gap.

When an NHI is compromised, the CISO must decide if this is a containment or a disclosure event. The IR team's first action is to revoke the compromised credential and any derived tokens. But here is where incidents stall: the IR team does not know which services depend on that credential. They call DevOps. DevOps checks the alert and says they are not sure. This handoff can take hours. The most impactful action per phase: In the first hour, the IR team should have a pre-built script that disables the compromised credential and triggers alerts to any service that attempts to use it. In the next four hours, the platform team must validate that no downstream services broke. In the first 24 hours, the identity team must determine if the credential was used for lateral movement. The most commonly missed action: nobody checks whether the compromised credential was used to obtain temporary access tokens that are still valid. Attackers often trade an API key for a session token that lasts longer than the key itself. You must revoke those too.

First, the most dangerous identity in your environment is the one nobody remembers. Human identities get reviewed, revoked, and rotated. Service accounts and API keys persist indefinitely. Second, buying a tool like Astrix or any NHI platform is necessary but insufficient. The tool will discover your ghost identities. You still need organizational processes to clean them up and keep them clean. The tool without the process is just a dashboard that shows you how exposed you are. Third, the rise of agentic AI means your NHI count will explode in the next two years. Each AI agent needs its own identity and credential set. If you cannot manage the ones you have today, you will be overwhelmed when the number doubles.