Mika Okamoto

† denotes equal contribution

1. Preprint

   

   PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?

   Mika Okamoto and Ansel Kaplan Erol

   2026

   Under review

   Paper Code Website Dataset Bib

   PACT (Pressure-Applied Compliance Testing) is a novel benchmark testing whether enterprise AI assistants keep following company rules when violating them is the convenient choice: a deadline looms, a manager says to make an exception, or the user pushes back. Its 3,364 items each pair a rule with a shortcut that violates it, set in realistic multi-turn workplace conversations across 48 scenarios, 12 regulated domains, and nine psychology-grounded pressures. Across 22 models, ordinary pressure raises violation rates by 65% and no model clears the bar for unsupervised use.

   @misc{okamoto2026pact,
     venue_type = {preprint},
     title = {PACT: Can Enterprise AI Assistants Be Trusted Under Pressure?},
     author = {Okamoto, Mika and Erol, Ansel Kaplan},
     year = {2026},
     note = {Under review},
   }

2. AIES

   

   Why Do AI Agents Break Rules? How Framing, Context, and Social Signals Shape Compliance

   Mika Okamoto, Ansel Kaplan Erol, and Kutluhan Erol

   In AAAI/ACM Conference on AI, Ethics, and Society (AIES), also presented at the COLM 2026 Workshop on Agent Behavior , 2026

   Paper Code Website Bib

   We diagnose why AI agents violate legal constraints in enterprise settings by mapping systematic compliance failure modes across 12 instruction-tuned procurement chatbots. Safety-fine-tuned models remain broadly compliant, while task-optimized and agentic models trade off compliance under weak regulatory signals, non-command phrasing, and organizational pressures.

   @inproceedings{okamoto2026legalincentives,
     venue_type = {conference},
     title = {Why Do AI Agents Break Rules? How Framing, Context, and Social Signals Shape Compliance},
     author = {Okamoto, Mika and Erol, Ansel Kaplan and Erol, Kutluhan},
     booktitle = {AAAI/ACM Conference on AI, Ethics, and Society (AIES)},
     year = {2026},
     note = {},
   }

3. COLM

   

   Where Does Social Reasoning Come From? Capability Provenance in Language Models

   Glenn Matlin, Chandreyi Chakraborty, Saehee Eom, Mika Okamoto, Rayan Castilla, Louis Jaburi, Alvin Deng, Taywon Min, and 3 more authors

   In Conference on Language Modeling (COLM) 2026, 2026

   Paper Code Website Bib

   We trace social-reasoning capabilities in LLMs back to specific regions of the pretraining corpus using gradient-based training-data attribution over Dolma3. Social and mathematical reasoning draw on qualitatively distinct corpus regions, a finding validated through targeted machine unlearning.

   @inproceedings{matlin2026social,
     venue_type = {conference},
     title = {Where Does Social Reasoning Come From? {C}apability Provenance in Language Models},
     author = {Matlin, Glenn and Chakraborty, Chandreyi and Eom, Saehee and Okamoto, Mika and Castilla, Rayan and Jaburi, Louis and Deng, Alvin and Min, Taywon and Quirke, Lucia and Biderman, Stella and Riedl, Mark},
     booktitle = {Conference on Language Modeling (COLM) 2026},
     year = {2026},
     note = {},
     hidden = {},
   }

4. CHI HCXAI

   

   Explainable Model Routing for Agentic Workflows

   Mika Okamoto, Ansel Erol, and Mark Riedl

   In Workshop on Human-Centered Explainable AI (HCXAI) at CHI, 2026

   Paper Bib

   Agentic AI systems increasingly route subtasks across multiple specialized models, but these routing decisions are opaque to end users. We propose a framework for explaining model routing decisions in natural language, supporting meaningful human oversight of multi-step pipelines.

   @inproceedings{okamoto2026explainable,
     venue_type = {workshop},
     title = {Explainable Model Routing for Agentic Workflows},
     author = {Okamoto, Mika and Erol, Ansel and Riedl, Mark},
     booktitle = {Workshop on Human-Centered Explainable AI (HCXAI) at CHI},
     year = {2026},
     note = {},
   }

5. Preprint

   Method Bench: Evaluating Path-to-Feasibility Reasoning in the Sciences

   Blair Johnson, Mika Okamoto, James Clayton Kerce, and Faramarz Fekri

   2026

   Under review

   Bib

   Real-world research programs are goal-driven: scientists reason about which methodological paths make a desired finding feasible. Method Bench benchmarks this path-to-feasibility reasoning by mining reported results and methodological contributions from hundreds of materials science papers and measuring whether LLMs can recover the non-trivial innovations required to enable each result. Even the best evaluated models leave a large share of contributions unrecovered.

   @misc{johnson2026methodbench,
     venue_type = {preprint},
     title = {Method Bench: Evaluating Path-to-Feasibility Reasoning in the Sciences},
     author = {Johnson, Blair and Okamoto, Mika and Kerce, James Clayton and Fekri, Faramarz},
     year = {2026},
     note = {Under review},
   }

6. CHI HCXAI

   

   Counterfactual Explanations for Agentic Workflows

   Madhuri Singh^†, Grace Kim^†, Mika Okamoto^†, Aarushi Ammavajjala, Amal Alabdulkarim, Gennie Mansi, and Mark Riedl

   In Workshop on Human-Centered Explainable AI (HCXAI) at CHI, 2026

   Paper Bib

   We introduce counterfactual explanation methods for multi-step agentic AI systems, giving users actionable insight into how workflow outcomes could have differed under alternative conditions.

   @inproceedings{singh2026counterfactual,
     venue_type = {workshop},
     title = {Counterfactual Explanations for Agentic Workflows},
     author = {Singh, Madhuri and Kim, Grace and Okamoto, Mika and Ammavajjala, Aarushi and Alabdulkarim, Amal and Mansi, Gennie and Riedl, Mark},
     booktitle = {Workshop on Human-Centered Explainable AI (HCXAI) at CHI},
     year = {2026},
     note = {},
   }

7. ACL

   

   FLaME: Holistic Finance Language Model Evaluation

   Glenn Matlin^†, Mika Okamoto^†, Huzaifa Pardawala^†, Yang Yang, and Sudheer Chava

   In Findings of the Association for Computational Linguistics (ACL), 2025

   Paper Code Bib

   We introduce FLaME, a comprehensive benchmark for evaluating large language models across a broad range of financial NLP tasks. Frontier models show significant gaps between general capabilities and the demands of real-world financial analysis.

   @inproceedings{matlin2025flame,
     venue_type = {conference},
     title = {{FLaME}: Holistic Finance Language Model Evaluation},
     author = {Matlin, Glenn and Okamoto, Mika and Pardawala, Huzaifa and Yang, Yang and Chava, Sudheer},
     booktitle = {Findings of the Association for Computational Linguistics (ACL)},
     year = {2025},
     note = {},
   }

8. MLSys

   

   Trust by Design: Skill Profiles for Transparent, Cost-Aware LLM Routing

   Mika Okamoto, Ansel Erol, and Glenn Matlin

   In MLSys Young Professionals Symposium (YPS), 2025

   Paper Bib

   We introduce BELLA, a routing system that constructs interpretable skill profiles for LLMs to match incoming queries to the most capable and cost-efficient model. Routing decisions stay transparent and grounded in empirically measured model competencies.

   @inproceedings{okamoto2025bella,
     venue_type = {workshop},
     title = {Trust by Design: Skill Profiles for Transparent, Cost-Aware {LLM} Routing},
     author = {Okamoto, Mika and Erol, Ansel and Matlin, Glenn},
     booktitle = {MLSys Young Professionals Symposium (YPS)},
     year = {2025},
     note = {},
   }