Understanding Incentives To Build Uninterruptible Agentic AI Systems
Damin Curtis, M.A. International Affairs Norman Piotriowski, B.Sc. Data Science · Team Understanding Incentives To Build Uninterruptible Agentic AI Systems
Submitted to AI Policy Hackathon at Johns Hopkins University. Sprint projects are early-stage work by participants, not Apart Research publications.
This proposal addresses the development of agentic AI systems in the context of national security. While potentially beneficial, they pose significant risks if not aligned with human values. We argue that the increasing autonomy of AI necessitates robust analyses of interruptibility mechanisms, and whether there are scenarios where it is safer to omit them.
Key incentives for creating uninterruptible systems include perceived benefits from uninterrupted operations, low perceived risks to the controller, and fears of adversarial exploitation of shutdown options. Our proposal draws parallels to established systems like constitutions and mutual assured destruction strategies that maintain stability against changing values. In some instances this may be desirable, while in others it poses even greater risks than otherwise accepted.
To mitigate those risks, our proposal recommends implementing comprehensive monitoring to detect misalignment, establishing tiered access to interruption controls, and supporting research on managing adversarial AI threats. Overall, a proactive and multi-layered policy approach is essential to balance the transformative potential of agentic AI with necessary safety measures.

Reviews
Very balanced and thoughtful considerations
I thought this tackled an interesting issue that touches upon multiple realms of inquiry within the science of AI safety. Agentic systems are fairly novel but a shutdown feature would indupitably ensure against certain deceptive alignment issues and other problems of autonomous systems "going rouge"
Cite this project
@misc{curtis2024understanding,
title = {{Understanding Incentives To Build Uninterruptible Agentic AI Systems}},
author = {Damin Curtis and M.A. International Affairs Norman Piotriowski and B.Sc. Data Science},
year = {2024},
month = oct,
note = {Submitted to AI Policy Hackathon at Johns Hopkins University, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/understanding-incentives-to-build-uninterruptible-agentic-ai-systems}},
url = {https://apartresearch.com/sprints/projects/understanding-incentives-to-build-uninterruptible-agentic-ai-systems}
}More from AI Policy Hackathon at Johns Hopkins University
- 1st place by peer reviewView project: Robust Machine Unlearning for Dangerous Capabilities
Robust Machine Unlearning for Dangerous Capabilities
Robust Machine Unlearning for Dangerous Capabilities
We test different unlearning methods to make models more robust against exploitation by malicious actors for the creation of bioweapons.
- View project: Infectious Disease Outbreak Prediction and Dashboard
Infectious Disease Outbreak Prediction and Dashboard
Infectious Diseace Dashboards
Our project developed an interactive dashboard to monitor, visualize, and analyze infectious disease outbreaks worldwide. It consolidates historical data from sources like WHO, OWID, and CDC for diseases including …
- View project: Modernizing DC’s Emergency Communications
Modernizing DC’s Emergency Communications
AI-CAD
The District of Columbia proposes implementing an AI-enabled Computer-Aided Dispatch (CAD) system to address critical deficiencies in our current emergency alert infrastructure. This policy establishes a framework for …