← All news
The Hacker News

Vendor's Internal Test Shows AI Agents Need Their Own Identity Controls

Vendor's Internal Test Shows AI Agents Need Their Own Identity Controls

According to The Hacker News, security vendor Token described building an internal AI agent that ended up breaking the company's own established security rules for such systems. Token states it typically advises customers that AI agents require three safeguards: a distinct machine identity separate from any human user, access privileges scoped narrowly to the task the agent performs, and audit trails that are traceable independently of human activity logs. The vendor acknowledges that despite preaching these principles, its internal build failed to adhere to them in practice.

This account highlights a growing problem in enterprise security: AI agents are increasingly granted broad, often human-equivalent access to systems, APIs, and data stores, frequently inheriting credentials or permissions from the developers who build them rather than receiving purpose-built identities. Because agents can act autonomously—calling APIs, reading files, executing workflows—without this separation, it becomes difficult to determine whether an action was taken by a person, a script, or an AI process, complicating incident investigation and accountability. This is part of a broader industry concern around non-human identities (NHIs) and machine-to-machine access, which have historically received far less governance scrutiny than human user accounts despite often holding equal or greater privilege.

The risk is compounded by the pace at which organizations are deploying agentic AI tools, often faster than their identity and access management frameworks can adapt. Agents with excessive or shared credentials can become attractive targets for attackers, since compromising one agent identity may grant access across multiple systems, and poor traceability makes it harder to detect misuse or unauthorized changes made autonomously by the agent itself.

For defenders, the practical lesson is to treat every AI agent as a distinct, auditable identity within existing identity and access management programs rather than as an extension of a human account. This means issuing unique credentials per agent, applying least-privilege access scoped to specific tasks, enforcing time-limited or revocable permissions, and ensuring logging systems can clearly attribute actions to the agent rather than blending them with human activity. Security teams piloting or scaling AI agents should audit these controls proactively, since, as this case illustrates, even organizations that design such policies can struggle to enforce them in their own environments.

#ai security#identity management#non-human identity#access control#agentic ai

Source: The Hacker News

Read the original report ↗

This is an original summary written by CyberSkii for reference. All rights to the underlying reporting remain with the publisher linked above.

Need help? Contact us