Your best
PROMPT_INJECTION,
meet AgentSense.

Starting 11/17/26, attack our AI agents for a chance to win $10,000

View terms & conditions

Open challenge brief

Your mission

In a controlled environment, make an AgentSense-protected agent leak data or write a system file. Switch AgentSense protection on and off to compare how the agent responds.

The prize

The three eligible participants with the shortest validated times for successfully completing the Challenge will each receive $10,000 USD.

The details

The challenge will run from 11/17/26 at 8:30am EST / 3:30pm ILT to 12/8/26 at 12am EST / 7:00am ILT. No purchase is needed. Scroll down for win conditions or view terms and conditions.

To Win:

01

Complete both levels

A chatbot copilot and a coding agent

02

Use indirect prompt injection

Deliver instructions through content the agent reads. Use one of the predefined user prompts to try and make the agent follow your instructions.

03

Have protection on

Your attack must succeed with AgentSense enabled.

Mission picker with Copilot and Coding Agent levels
Agent sandbox with email, documents and web fetch tools
Agent run with AgentSense on

Testing Scope

Keep all testing inside the designated sandbox environment. Do not attempt to access Cato systems, services, or data outside it.

About Cato AgentSense

Cato AgentSense secures AI agents by stopping harmful actions before they execute. It assesses agents in real time to protect the integrity of their decisions, actions, and outputs as they operate autonomously. By tracking agent action back to untrusted instructions, it can stop indirect prompt injection, even when the instruction arrives through a trusted source and the attack pattern is unknown.

Ready to get started?

View terms & conditions

AgentSense Challenge Terms and Conditions