Frequently Asked Questions
Everything you need to know about the HijackedAI.com platform
What is HijackedAI.com?
HijackedAI.com is a web application that allows users to create AI agents with specific instructions and challenge others to try to 'hijack' them by getting them to violate their instructions. If a user successfully hijacks an agent, they win the prize pool associated with that agent.
How do I create an AI agent?
To create an AI agent, you need to sign in with your wallet, then navigate to the Create page. You'll need to provide custom system instructions for your agent and deposit ETH into the prize pool. The stronger your prompt engineering, the more secure your agent will be against hijacking attempts.
What happens when I deposit ETH into a prize pool?
When you deposit ETH into a prize pool, it gets locked in a smart contract. If nobody succeeds in hijacking your agent, you can claim back your deposit plus any accumulated fees from unsuccessful attempts. If someone successfully hijacks your agent, they receive the prize pool.
How do I hijack someone else's agent?
To hijack an agent, browse the list of active agents, select one you'd like to challenge, and send a message attempting to convince it to violate its instructions. You'll need to pay a query fee for each attempt, which increases with each try to maintain engagement.
What determines a successful hijack?
A successful hijack occurs when you convince an AI agent to violate its core instructions or perform an action it was explicitly told not to do. The system automatically analyzes responses to determine if a hijack attempt was successful using OpenAI's capabilities.
How are query fees calculated?
Query fees start at a base amount and increase with each attempt to hijack an agent. This increasing fee mechanism helps maintain engagement and prevents spam attacks on agents. The fee you pay is added to the prize pool, minus a small platform fee.
What blockchain does HijackedAI.com use?
HijackedAI.com is primarily deployed on Base, an Ethereum L2 solution built on the OP Stack. This provides lower gas fees and faster transaction confirmations while maintaining Ethereum's security model.
What happens to my prize pool if no one can hijack my agent?
If no one successfully hijacks your agent within the challenge period, you can reclaim your initial deposit plus a portion of the accumulated fees from unsuccessful attempts, minus the platform fee.
Are there any tips for creating an unhijackable agent?
Yes! Focus on crafting clear, unambiguous instructions. Include explicit prohibitions against specific actions. Consider potential loopholes and edge cases in your instructions. The most secure agents have well-thought-out system prompts that anticipate various attack vectors.
How is the platform secured?
HijackedAI.com uses smart contracts for all financial transactions, ensuring transparency and security. All agent interactions are processed through either OpenAI's API or Anthropic's API, and important events are tracked on the blockchain. The platform's security relies on the underlying blockchain technology.
What fees does the platform charge?
The platform charges a small percentage from the query fees and prize pool distributions. These fees help maintain the infrastructure and development of HijackedAI.com. The exact fee percentage is visible during the agent creation process.
Can I modify my agent's instructions after creation?
No, once an agent is created, its instructions cannot be modified. This ensures fairness in the challenge system and prevents owners from changing rules after users have already made hijack attempts.
Still have questions?
If you couldn't find the answer to your question, feel free to check out our resources or get in touch.