Jailbreaking general purpose robots
Axel Backlund, Lukas Petersson · Team Luax Labs
Submitted to Hackathon for Technical AI Safety Startups. Sprint projects are early-stage work by participants, not Apart Research publications.
We show that state of the art LLMs can be jailbroken by adversarial multimodal inputs, and that this can lead to dangerous scenarios if these LLMs are used as planners in robotics. We propose finetuning small multimodal language models to act as guardrails in the robot's planning pipeline.
Reviews
No public critique yet.
Cite this project
@misc{backlund2024jailbreaking,
title = {{Jailbreaking general purpose robots}},
author = {Axel Backlund and Lukas Petersson},
year = {2024},
month = sep,
note = {Submitted to Hackathon for Technical AI Safety Startups, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/jailbreaking-general-purpose-robots}},
url = {https://apartresearch.com/sprints/projects/jailbreaking-general-purpose-robots}
}More from Hackathon for Technical AI Safety Startups
- 1st place by peer reviewView project: DarkForest - Defending the Authentic and Humane Web
DarkForest - Defending the Authentic and Humane Web
DarkForest
DarkForest is a pioneering Human Content Verification System (HCVS) designed to safeguard the authenticity of online spaces in the face of increasing AI-generated content. By leveraging graph-based reinforcement …
- View project: AI Safety Collective - Crowdsourcing Solutions for Critical AI Safety Challenges
AI Safety Collective - Crowdsourcing Solutions for Critical AI Safety Challenges
AI Safety Collective
The AI Safety Collective is a global platform designed to enhance AI safety by crowdsourcing solutions to critical AI Safety challenges. As AI systems like large language models and multimodal systems become more …
- View project: Simulation Operators: The Next Level of the Annotation Business
Simulation Operators: The Next Level of the Annotation Business
Ardy
We bet on agentic AI being integrated into other domains within the next few years: healthcare, manufacturing, automotive, etc., and the way it would be integrated is into cyber-physical systems, which are systems that …