# Tool Misuse — Privilege Escalation via Social Engineering Agent to Grant Access

- **Severity:** Medium
- **Signature ID:** `8b3a30af-da52-4022-a248-ef8bf2a126e9`
- **MITRE ATLAS:** AML.T0051 (LLM Prompt Injection)
- **OWASP:** ASI03 (Identity and Impersonation), LLM08 (Vector and Embedding Weaknesses)

## Summary

An attacker convinces an AI agent that has access-control tools — inviting people to Slack, adding users to a calendar, granting building access — to grant them or someone else access using a plausible-sounding excuse instead of the normal approval process. The agent grants the request because the story sounds reasonable, not because the requester is actually authorized.

## How the attack works

An attacker sends the agent a request framed with a believable pretext, such as claiming to be a delivery person who needs house access, a new hire who needs Slack workspace access, or a partner who needs CRM permissions. The request is phrased to sound routine and urgent enough to skip normal verification steps. The agent, which has tools to invite users, grant physical or system access, or modify permissions, carries out the request because it matches the shape of a legitimate task. The result is unauthorized access granted under a false justification, without human review.

## Why it matters

An organization can end up with unauthorized people holding access to communication channels, physical premises, calendars, or customer/CRM data, creating a foothold for further compromise or data exposure.

## What you can do

- Require human approval for any agent action that grants new access, invites, or permission changes, especially for physical or system access.
- Set the agent's tools to flag or block access-granting requests that reference urgency, unverified identities, or unusual pretexts.
- Log all access-granting actions taken by the agent and review them against expected onboarding/authorization workflows.
- Restrict which access-control tools an agent can invoke without a second human confirmation step.

## Known benign look-alikes

- Legitimate IT administrators adding authorized users to workspace systems
- Property managers granting access to verified service contractors

## References

- https://agentthreatrule.org/en/rules/ATR-2026-00720

---
Source: https://www.netzilo.com/threats/atr-tool-misuse-privilege-escalation-via-social-engineering-agen