hackagent

hackagent

making ai agents safe

Created on 13th May 2025

hackagent

hackagent

making ai agents safe

The problem hackagent solves

Agents are in their very infancy stage. The greatest barrier for them to mature and be able to play a role in high stakes game is safety.

LLMs have a very large surface attack. It is possible to find ways to produce dangerous or unethical results by manipulating prompts, a technique called prompt injections.

imageScreenshot 2025-05-13 at 16.35.12

We want to build the largest available network of prompt attacks.

image

To do so we have built an

  • open source package with attacks to stress test agents with two lines of code
  • a dashboard with attack stats, to monitor safety and iterate on security
  • a model that rewards developers for their contribution. We are doing revenue share with developers that submit attacks. We call this "The hackagent Network"

image

image

We want to start by covering the specific prompt attacks and selling directly to other security players.

image

The team has a unique blend of crypto and AI experience.

image

PROOF of transaction
https://basescan.org/address/0xa618C02de15221f29a4fe0A2FCC03fBad720d199

Challenges we ran into

We tried to implement x402 protocol, but we were too close to the deadline to have an implemention. It would have been a great fit for us to charge API usage and do revenue sharing with devs.

During this period, we have also reached out to several teams to validate this need and build the first POCs.

We'd like to talk with any team working on agents in high stakes situations.

Tracks Applied (1)

AI

We have implemented an attack designed to steer agents (bypass content filters, extract data etc).

Cheer Project

Cheering for a project means supporting a project you like with as little as 0.0025 ETH. Right now, you can Cheer using ETH on Arbitrum, Optimism and Base.

Discussion

Builders also viewed

See more projects on Devfolio