.png){kind=icon}
.png){kind=icon}
Return
An agentic AI marketplace for prelitigation analyses with ZKP-integrated ethical verifications
.png){kind=icon}
PRE
AI
U
D
B
K
刘
DOI：10.3389/fbloc.2026.1770848.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
An agentic AI system could assist in reducing case workloads in the judiciary by providing a probable outcome for litigating parties and eliminating the need to go to court. However, agentic AI uses different ML and LLM models in its workflows, and these models may reflect biases present in their training data. These biases could affect the outcome of the pre-litigation proceedings, favouring one litigating party over the other, depending on demographic representations in the training dataset. We consider an Agentic AI marketplace where litigants can verify demographic representation in the training data of ML/LLM models used in each Agentic AI before selecting a suitable system for pre-trial analyses. ZKPs provide cryptographic primitives for verifying such claims without disclosing full information about the underlying dataset. This paper outlines a conceptual framework for performing these verifications within the European regulatory context. First, we demonstrate the implementation of an Agentic AI system for prelitigation analyses and then conceptualize an Agentic AI marketplace, outlining different technological interactions. We then formalise demographic representation as a verifiable property and outline a ZKP framework for verification, comprising different tech stacks like BBS+ signatures, bulletproofs, zk-SNARKs, and smart contract oracles. This paper presents a conceptual and architectural framework rather than an empirical system evaluation, and aims to establish a foundational design space for privacy-preserving bias verification in agentic legal AI.
Keywords:
agentic AI
AI ethics
bias verification
ML/LLM
prelitigation analyses
ZKPs
Journal
F
IF:
2.4
Papers:
59
Citations:
558
Organization
Cited Papers