PROV-AGENT: Unified Provenance for Tracking AI Agent Interactions in Agentic Workflows
Renan Souza 1 , Amal Gueroudji 2 , Stephen DeWitt 3 , Daniel Rosendo 1 , Tirthankar Ghosal 1 , Robert Ross 2 , Prasanna Balaprakash 4 , Rafael Ferreira da Silva 1 · 2025
Read PROV-AGENT: Unified Provenance for Tracking AI Agent Interactions in Agentic Workflows by Renan Souza 1 , Amal Gueroudji 2 , Stephen DeWitt 3 , Daniel Rosendo 1 , Tirthankar Ghosal 1 , Robert Ross 2 , Prasanna Balaprakash 4 , Rafael Ferreira da Silva 1 (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-02866.txt.
Description
Souza and seven co-authors set out PROV-AGENT, a way to keep a full trail of what AI agents do inside shared work flows, in work from 2025. Agents plan tasks and hand their work to the next agent, across sites that stay apart on mixed gear. An agent can make up an output, and that flaw can spread in the dark as the next agent takes it as input. To see and trace each step thus counts a lot when teams judge trust.
The model builds on the W3C PROV norm and adds tags about each agent step. Prompts, replies, and choices are tied to the wider flow and to what came of them. An open tool grabs that trail live as work runs. Tests span edge sites, cloud sites, and big lab machines. Key trail queries and trust checks work across all of them.
Catalog record
| CREATOR | Renan Souza 1 , Amal Gueroudji 2 , Stephen DeWitt 3 , Daniel Rosendo 1 , Tirthankar Ghosal 1 , Robert Ross 2 , Prasanna Balaprakash 4 , Rafael Ferreira da Silva 1 |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-02866