Using ARC-AGI puzzles as CAPTCHa task
Mikolaj Kniejski · Team Kapcza
Submitted to Agent Security Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
This project has no abstract. The report has the details.
Reviews
This is more of a concept submission without any PoC implementation. The presentation is not detailed enough to explore concerns around AI getting better at ARC puzzles in the future etc. This is more of a temporary conceptual solution without test results.
CAPTCHA is an important and very relevant problem to solve from a security perspective, especially in current world where LLM agents can potentially solve >50% of the deployed CAPTCHAS. However, I wonder if ARC AGI puzzles maybe too challenging for a non-expert human. As such, such security measures can hamper the ability of a genuine non expert user to access a system.
Cite this project
@misc{kniejski2024using,
title = {{Using ARC-AGI puzzles as CAPTCHa task}},
author = {Mikolaj Kniejski},
year = {2024},
month = oct,
note = {Submitted to Agent Security Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/using-arc-agi-puzzles-as-captcha-task}},
url = {https://apartresearch.com/sprints/projects/using-arc-agi-puzzles-as-captcha-task}
}More from Agent Security Hackathon
- 1st place by peer reviewView project: Diamonds are Not All You Need
Diamonds are Not All You Need
Diamonds are Not All You Need
This project tests an AI agent in a straightforward alignment problem. The agent is given creative freedom within a Minecraft world and is tasked with transforming a 100x100 radius of the world into diamond. It is …
- View project: Cross-model surveillance for emails handling
Cross-model surveillance for emails handling
Fluffy Vin
A system that implements cross-model security checks, where one AI agent (Agent A) interacts with another (Agent B) to ensure that potentially harmful actions are caught and mitigated before they can be executed. …
- View project: Inference-Time Agent Security
Inference-Time Agent Security
Inference-Time Agent Security
We take a first step towards automating model building for symbolic checking (eg formal verification, PDDL) of LLM systems.