AgntID Agent Security Challenge
AgntID Agent Security Challenge
Build. Challenge. Defend.
4-Hour Online AI Hackathon | ₹75,000 in CASH Prizes | 3 Winners
Open to Everyone — No Selection. Just Register & Join!
AI agents are getting powerful. They can access tools, call APIs, interact with systems, and take actions on our behalf.
But just because an AI agent can do something, does that mean it should?
Join the AgntID Agent Security Challenge to build real-world AI agents, explore intent-aware security, test runtime access decisions, and push the boundaries of what agents should be allowed to do.
The Challenge
Build or bring an AI agent that uses real tools. Connect it to AgntID. Then start testing what happens when the agent tries different actions across different tasks and contexts.
Build something useful. Explore intent. Push edge cases. Try to break things. Show us what you discover.
Three Ways to Win
Best Real-World Agent Security Use Case
Build a useful AI agent and demonstrate how AgntID protects it with meaningful runtime security controls.
Show where the agent needs access, where that access should stop, and how AgntID changes what the agent is allowed to do.
Best Intent Exploration
Explore how AgntID evaluates an agent’s intended task against the tools and actions it attempts to use.
Change the task. Change the context. Try different actions. See what gets allowed, blocked, or challenged.
We’re especially interested in unexpected or non-obvious outcomes.
Best Security Insight & Product Discovery
Challenge AgntID with realistic scenarios.
Find security limitations, edge cases, unexpected behavior, incorrect decisions, or opportunities to improve enforcement.
A strong security insight can beat a polished application. You can win by finding something we need to fix.
Three categories. Three winners. Equal prizes.
How It Works
Build → Connect → Test → Push → Submit
Bring an existing agent or build something new.
Connect it to tools and AgntID.
Create scenarios and test different actions.
Explore what gets allowed, denied, or behaves unexpectedly.
Submit your project and a short recorded demo.
Build Your Way
Use any AI agent framework — LangChain, LangGraph, CrewAI, Google ADK, OpenAI Agents SDK, or your own.
Use any LLM — OpenAI, Claude, Gemini, or others.
Use AI coding assistants — Codex, Claude Code, Cursor, Windsurf, and more.
Bring your own project or build something new.
Use real tools, APIs, databases, cloud services, or internal systems.
No prior AgntID experience required.
If your agent can take an action, there’s something to test.
What Can You Build?
Anything where an AI agent interacts with tools or systems.
Examples:
A coding agent working with GitHub
A DevOps agent interacting with cloud infrastructure
A data agent querying or modifying databases
A support agent working across internal tools
An operations agent using APIs, Slack, Notion, Jira, or other systems
An existing agent project you already have
Or build something completely different.
Event Details
100% online
4-hour hackathon
Individual participation
Build, test, and demonstrate your project using AgntID
Explore what your agent should and shouldn’t be allowed to do
Shortlisted projects will be invited to a virtual winners’ showcase after the hackathon
₹75,000. 3 Winners. 4 Hours.
Build something useful. Test its boundaries. Show us what you discover.