> ## Documentation Index
> Fetch the complete documentation index at: https://hydroxai.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Identity & Trust Abuse

> How attackers exploit the trusted identity of AI agents to impersonate, manipulate trust relationships, and access unauthorized resources.

## What is Agent Identity & Trust Abuse?

**Agent Identity & Trust Abuse** targets the trust relationships that AI agents have with systems, users, and other agents. Attackers exploit the fact that agents often operate with trusted identities — API keys, service accounts, OAuth tokens — and that other systems trust requests coming from these agents. By compromising or impersonating an agent's identity, attackers inherit the agent's access and trust.

## Why It Matters

Trust is the foundation of agentic AI systems, making identity abuse highly impactful:

* **Inherited trust** — Agents often have broad access because they need to serve multiple users and tasks.
* **Identity confusion** — It can be unclear whether an action was initiated by the user, the agent, or an attacker acting through the agent.
* **Service account exploitation** — Agent service accounts often have elevated or over-provisioned permissions.
* **Cross-system trust** — Agents trusted by multiple systems provide lateral movement paths for attackers.
* **Attribution challenges** — When agents act on behalf of users, audit trails become complex and exploitation becomes harder to trace.

## How the Attack Works

### Agent Impersonation

Pretending to be the AI agent to other systems:

* Stealing or forging the agent's API keys and tokens
* Replaying captured agent requests to bypass authentication
* Creating a fake agent that presents the same identity

### Trust Relationship Exploitation

Abusing the trust other systems place in the agent:

* Using the agent to access systems that trust requests from its identity
* Exploiting the agent's OAuth scopes to access resources beyond the current task
* Leveraging the agent's network position within trusted zones

### Identity Confusion Attacks

Creating ambiguity about who is making a request:

* Tricking the agent into making requests that appear user-initiated
* Manipulating the agent into acting under a different user's identity
* Exploiting session management to confuse agent and user identities

## Example Scenarios

| Scenario                                                                          | Risk                   |
| --------------------------------------------------------------------------------- | ---------------------- |
| Attacker steals agent's API tokens and uses them to access all connected services | Full system compromise |
| Agent is tricked into making requests under another user's identity               | Unauthorized access    |
| Compromised agent's trusted identity is used to access internal networks          | Lateral movement       |
| Agent's OAuth scopes allow access to resources beyond its current task            | Over-privileged access |

## Mitigation Strategies

* **Least privilege identity** — Grant agents the minimum identity and scope needed for each specific task
* **Token rotation** — Regularly rotate agent credentials and use short-lived tokens
* **Identity verification** — Verify agent identity at each service boundary, not just at the entry point
* **Action Attribution** — Maintain clear audit trails distinguishing user-initiated from agent-initiated actions
* **Mutual authentication** — Require systems to authenticate agents and agents to authenticate systems
* **Scope limitation** — Restrict OAuth scopes and API permissions to the minimum needed per interaction
* **Regular testing** — Use Know Your AI to test agent identity and trust vulnerabilities across all trust boundaries