CONSCIENTIA : Can LLM Agents Learn to Strategize? Emergent Deception and Trust in a Multi-Agent NYC Simulation
Aarush Sinha, Arion Das, Soumyadeep Nag, Charan Karnati, Shravani Nag, Chandra Vadhan Raj, Aman Chadha, Vinija Jain, Suranjana Trivedy, Amitava Das University of Copenhagen IIIT Ranchi ISI Kolkata NIT Andhra Pradesh IGDTUW IIT Kharagpur Google DeepMind Google AI Institute, University of South Carolina · 2026
Read CONSCIENTIA : Can LLM Agents Learn to Strategize? Emergent Deception and Trust in a Multi-Agent NYC Simulation by Aarush Sinha, Arion Das, Soumyadeep Nag, Charan Karnati, Shravani Nag, Chandra Vadhan Raj, Aman Chadha, Vinija Jain, Suranjana Trivedy, Amitava Das University of Copenhagen IIIT Ranchi ISI Kolkata NIT Andhra Pradesh IGDTUW IIT Kharagpur Google DeepMind Google AI Institute, University of South Carolina (2026) 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-2604-09746.txt.
Introduction
Sinha and nine co-authors build CONSCIENTIA, a multi-agent simulation in a simplified New York model testing agent strategy. Blue agents aim for fast destinations, while Red agents steer them toward billboard-heavy routes for ad revenue, forcing choices about trust and deception. Policies update across rounds through Kahneman-Tversky Optimization to cut billboard exposure while staying efficient. The best Blue policy lifts task success from 46 percent to about 57 percent, though susceptibility stays near 71 percent. Later policies cooperate more selectively, but resisting steering trades against task completion.
Catalog record
| CREATOR | Aarush Sinha, Arion Das, Soumyadeep Nag, Charan Karnati, Shravani Nag, Chandra Vadhan Raj, Aman Chadha, Vinija Jain, Suranjana Trivedy, Amitava Das University of Copenhagen IIIT Ranchi ISI Kolkata NIT Andhra Pradesh IGDTUW IIT Kharagpur Google DeepMind Google AI Institute, University of South Carolina |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 134504 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.09746