Lisa Intel: A Global Runtime Governance Layer for Autonomous AI

AI systems make more and more qualified and complex desicions, take own actions and interact in the real world. Still, most AI safety solutions today focus on individual models or agents, I believe that we also need independent safety solutions. In part because we cannot trust to 100% that the companies building these models and agents always will succed with their safety and in part because many builders of models and systems might not care about safety, especially if they for instance are rogue actors.

This project aims to develop a model-agnostic runtime governance platform/system, which operates between AI systems and the environments where they act. The system is designed to authorize actions, allow human oversight, maintain auditability and support accountable deployment across different AI models and organizations.

The long term vision for my system is that it would become a global governance layer that can coordinate safety in the use of advanced AI models working in different sectors and jurisdictions, independent of who the provider is. A simple way to describe the idea is comparing it to the role of the common internet protocols when the internet started.

During the project time I will continue to develop the system, carry out adversarial testing, improve the runtime governance capabilities and validate work and approach with researchers and enterprise partners. The goal being a working research prototype which demonstrates how advanced autonomous AI systems can work under independent, external governance.

The funding would allow me to focus full time on building Lisa Intel. It will be used for development, cloud infrastructure, testing, adversarial evaluations, patent costs related to the project and basic living expenses.

My goal is to deliver a working research prototype and demonstrate that independent runtime governance can make advanced autonomous AI systems safer to deploy.