Can key existing accepted business and legal frameworks for web-based transactions and intermediaries be operationalized at the interface with agentic AI systems in a way that avoids or mitigates risk, ensures accountability, and yields predictable legal outcomes when harms arise?
Hypothesis 1: Key provisions of existing law governing use of electronic agents and autonomous transactions (such as UETA, ESIGN, and other relevant legal frameworks) can easily be leveraged to directly address many of the crucial challenged presented by use of agentic AI systems.
Hypothesis 2: Focusing the scope of this research on scenarios and use cases involving an agentic AI system conducting any form of transaction between a user-principal and a third party reflects many existing and near term situations and unlocks the appliction of the aforementioned legal frameworks.
Hypothesis 3: A standardized agent interface protocol, informed by the principles of the aforementioned legal frameworks, is a simple way to effectuate attribution of actions to legal parties, facilitating accountability when harms occur, and the learning from such a protocol can inform other technical approaches for achieving the same business and legal results.
Our research will leverage the robust legal frameworks of UETA, ESIGN, and other existing legal frameworks to construct a comprehensive operational model for agentic AI systems engaged in transactions. The core scenario involves transactions conducted by an agentic system on behalf of a user-principal with third parties. We'll explore variations where different actors, like service providers, are part of the transactional chain.
Daniel "Dazza" Greenwood
Alex "Sandy" Pentland