A Proposed Open Manifesto for AI Agents Touching Production Systems (aiagentmanifesto.org)

A new Open Manifesto has been proposed to guide the development of AI agents that interact with production systems, marking a pivotal shift from traditional deterministic automation to advanced probabilistic autonomy. This manifesto outlines essential principles aimed at mitigating risks associated with AI's unpredictable nature, especially concerning 'Stochastic Liability.' Key principles include the separation of reasoning and action, ensuring that AI does not execute decisions in a single opaque step, and the necessity for every output to possess a detailed creation history to establish accountability. The significance of this manifesto lies in its potential to foster trust in autonomous systems crucial for critical infrastructure and high-stakes environments. By emphasizing safety mechanisms like external oversight, immutable evidence, and episodic identity for agents, the manifesto seeks to enhance the reliability of AI workflows while mitigating catastrophic risks. Adopting these principles could lead to a more robust framework for deploying autonomous technologies, promoting a culture of verification and safety that aligns with the rapid evolution of AI/ML capabilities.