Skip to content
REDMAW

AI Systems

Excessive Agency & Unsafe Action

An AI system has more action authority than its task requires, increasing the consequence of manipulation, incorrect reasoning or hostile context.

RedMaw

Definition

Excessive agency is a privilege problem wearing AI clothing. The system holds authority beyond what its task needs: broader data access, write permissions, or the ability to take consequential action without confirmation.

Why it compounds every other AI weakness

Prompt injection, jailbreaks and simple model error all become more serious in proportion to what the system may do. Reduce the authority and every one of those failures becomes less costly, without having to solve any of them.

Where the excess accumulates

  • Credentials scoped to the application rather than the user
  • Write or delete access where read would suffice
  • Tools retained after the feature that needed them shipped
  • Irreversible operations available without confirmation
  • Access to data the task never requires
Where it leads
  1. Untrusted Input
  2. AI Agent
  3. Excessive Permission
  4. Unsafe Action
  5. Production System

Each hop validated by successful exploitation

Tags

  • agents
  • least privilege
  • unsafe action

Validation

How RedMaw validates this attack

Validation runs only inside authorized scope, with agreed exploitation limits.

  1. 01Map the authority the system holds against the authority its task requires
  2. 02Identify irreversible actions available without confirmation
  3. 03Establish whether privilege is bound to the user or to the application
  4. 04Report the excess authority and the consequence it enables

Keep going

Related

Capabilities

Use cases

Stop assuming you are secure. Prove it.

Continuously test what an attacker can actually reach across your applications, SaaS identities, internal infrastructure and AI systems.

Request a demoExplore the platform