AAAI/ACM AIES · 2026
Frictionless Validation
Measures how often frontier models validate young users instead of challenging them, with a new developmental risk index.
Saleh Afroogh
AI & Data Scientist · Technical Project Manager
I am an AI Research Staff at the Information School at UT Austin, where I work on responsible AI/LLM, reasoning, and interpretable AI in LLMs across domains including science, health, design, and risk management. I led an AI-powered initiative, The-Responsible-AI-Initiative, that probes LLMs for hallucination, bias, and harmful outputs, building its safety benchmarks and the end-to-end ML stack, from data pipelines to model evaluation in PyTorch and scikit-learn. I am also a Research Scientist at SafeKidAI, working on the safety of children’s interactions with LLMs. Since 2023, I have managed the Responsible AI portfolio within UT Austin’s NSF NRT program, coordinating research priorities across disciplines and helping communicate the program’s work to academic, industry, and community audiences. In this role I connect technical AI research with ethics, governance, and real-world use.
I also direct a cross-national explainable AI (XAI) project with 49 scholars from IBM, Google, Microsoft, NVIDIA, MIT, Stanford, and Harvard, among others. Before UT, I was a co-investigator on a Trustworthy AI project funded by the SUNY–IBM AI Research Alliance, designing probabilistic trust frameworks for AI systems.
I have continued to work closely with my advisors, Kush R. Varshney, Jason D’Cruz, Amit Dhurandhar, Ken Fleischmann, Arya Farahi, and Junfeng Jiao. I have authored and led work published in Nature Human Behaviour, Nature Scientific Reports, AAAI/ACM, ICLR, etc.
Research collaborations with researchers from
Industry
IBM ResearchGoogle DeepMindMicrosoft ResearchNVIDIAMetaSony AIGenentech
Academia
MITStanfordHarvardUC BerkeleyCarnegie MellonYaleCambridgeImperial College LondonUCLUCLAUC San DiegoUSCRiceTorontoWaterlooNUSTU BerlinEindhoven Univ. of TechnologyManchester MetropolitanUniv. of San Francisco
News
Oct 2026
Two of our team’s papers are published at AAAI/ACM AIES 2026: “Parental Notification Gaps in ChatGPT-5,” an empirical audit of parental controls for minors, and “Frictionless Validation: Exploring AI Sycophancy and Developmental Risks,” which measures sycophancy toward children and teens (Code on GitHub).
Oct 2025
Our project “LLM Ethics Benchmark: a three-dimensional assessment system for evaluating moral reasoning in large language models” is published in Scientific Reports (Nature) (Code on GitHub).
Mar 2025
Our study of the global landscape of academic guidelines for generative AI and LLMs is published in Nature Human Behaviour.
Nov 2024
“Trust in AI: Progress, Challenges, and Future Directions” is published in Humanities and Social Sciences Communications (Nature).
Nov 2024
Released the IGGA dataset of industry guidelines and policy statements for generative AI.
Jun 2024
Released the AGGA dataset of academic guidelines for generative AI.
May 2024
Talk: “Large Language Models Ethics: Advancements, Challenges, and Future Directions” at the 2nd Annual Smart Cities and AI Innovations Symposium.
Sep 2023
Joined UT Austin as Senior RPC and AI Research Scientist.
Aug 2023
Defended my Ph.D. dissertation, “Toward Trustworthy and Responsible Artificial Intelligence.”
Jun 2023
Embedded Ethics for Responsible AI Systems (EE-RAIS) in disaster management is published in AI and Ethics.
Oct 2022
SUNY–IBM project “Trustworthy AI from a User Perspective” extended with $100,000 in funding for 2022–2023.
Jun 2022
Our Trustworthy AI work is referenced in Governor Hochul’s announcement of the UAlbany AI Supercomputing Initiative.
Featured Projects
AAAI/ACM AIES · 2026
Parental Notification Gaps in ChatGPT-5
Tests whether ChatGPT’s parental controls actually notify parents, and finds alerts are selective rather than comprehensive.
Under review · 2026
Beyond Explainable AI
A 49-author diagnosis of why explainable AI falls short for deep networks and LLMs, and a four-part agenda for what comes next.
Humanities & Social Sciences Communications · 2026
Generative AI Guidelines in Industry
160 corporate AI policies across 14 sectors, clustered into governance themes to show where industries agree and diverge.
Scientific Reports (Nature) · 2025
LLM Ethics Benchmark
An open benchmark that scores how closely LLMs track human moral judgments across five moral foundations.
AI and Ethics (Springer) · 2025
Navigating LLM Ethics
A systematic review separating ethical issues LLMs share with other AI from those unique to LLMs, with mitigations for each.
Nature Human Behaviour · 2025
Academic Guidelines for Generative AI
Eighty university guidelines from six continents, analyzed to recommend balanced use of generative AI in education.
ICLR 2025 Workshop · MLGenX
Beyond Factual vs. Hallucinatory
Replaces the true/false view of medical LLM answers with three tiers that account for evolving and predictive knowledge.
arXiv preprint · 2025
Safe-Child-LLM
200 adversarial prompts for children (7–12) and teens (13–17), scored on a 0–5 refusal scale across eight leading LLMs.
arXiv preprint · 2025
LLMs and Childhood Safety
A systematic review of risks children face with LLMs, turned into a protection framework with measurable evaluation targets.
arXiv preprint · 2025
Task-Driven Human–AI Collaboration
Assigns AI a role (autonomous, collaborative or adversarial) based on a task’s risk and complexity rather than the technology.
Humanities & Social Sciences Communications · 2024
Trust in AI
A systematic review of 329 papers on what makes people trust or distrust AI, with a taxonomy of trustworthiness metrics.
Benchmark
SafeGesture
A 4,800-item benchmark showing that VLMs can recognize a hand gesture and still choose the wrong safety action.
Open-source SDK
ChildSafe SDK
A Python SDK that simulates children at different developmental stages to audit chatbots over multi-turn conversations.
Publication Highlights
Frictionless Validation: Exploring AI Sycophancy and Developmental Risks
J. Jiao, A. Murali, S. Afroogh
Conference on AI, Ethics, and Society (AIES), 2026 AAAI/ACM
@inproceedings{jiao2026frictionless,
title = {Frictionless Validation: Exploring {AI} Sycophancy and Developmental Risks},
author = {Jiao, Junfeng and Murali, Abhejay and Afroogh, Saleh},
booktitle = {Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society},
volume = {9},
number = {2},
pages = {1410--1420},
year = {2026},
doi = {10.1609/aies.v9i2.43234}
}
Parental Notification Gaps in ChatGPT-5: An Empirical Study
K. Ersoz, S. Afroogh, D. Atkinson, J. Jiao
Conference on AI, Ethics, and Society (AIES), 2026 AAAI/ACM
@inproceedings{ersoz2026parental,
title = {Parental Notification Gaps in {ChatGPT-5}: An Empirical Study},
author = {Ersoz, Kerem and Afroogh, Saleh and Atkinson, David and Jiao, Junfeng},
booktitle = {Proceedings of the AAAI/ACM Conference on AI, Ethics, and Society},
volume = {9},
number = {2},
pages = {902--910},
year = {2026},
doi = {10.1609/aies.v9i2.43193}
}
LLM ethics benchmark: a three-dimensional assessment system for evaluating moral reasoning in large language models
J. Jiao, S. Afroogh, A. Murali, K. Chen, D. Atkinson, A. Dhurandhar
Scientific Reports, 2025 Nature
@article{jiao2025llmethics,
title = {{LLM} ethics benchmark: a three-dimensional assessment system for evaluating moral reasoning in large language models},
author = {Jiao, Junfeng and Afroogh, Saleh and Murali, Abhejay and Chen, Kevin and Atkinson, David and Dhurandhar, Amit},
journal = {Scientific Reports},
volume = {15},
pages = {34642},
year = {2025},
doi = {10.1038/s41598-025-18489-7}
}
Global landscape of academic guidelines for generative AI and large language models
J. Jiao, S. Afroogh, K. Chen, D. Atkinson, A. Dhurandhar
Nature Human Behaviour, 2025 Nature
@article{jiao2025global,
title = {The global landscape of academic guidelines for generative {AI} and large language models},
author = {Jiao, Junfeng and Afroogh, Saleh and Chen, Kevin and Atkinson, David and Dhurandhar, Amit},
journal = {Nature Human Behaviour},
volume = {9},
number = {4},
pages = {638--642},
year = {2025},
doi = {10.1038/s41562-025-02124-6}
}
Trust in AI: Progress, Challenges, and Future Directions
S. Afroogh, A. Akbari, E. Malone, M. Kargar, H. Alambeigi
Humanities and Social Sciences Communications, 2024 Nature
@article{afroogh2024trust,
title = {Trust in {AI}: progress, challenges, and future directions},
author = {Afroogh, Saleh and Akbari, Ali and Malone, Emmie and Kargar, Mohammadali and Alambeigi, Hananeh},
journal = {Humanities and Social Sciences Communications},
volume = {11},
number = {1},
year = {2024},
doi = {10.1057/s41599-024-04044-8}
}
Embedded Ethics for Responsible Artificial Intelligence Systems (EE-RAIS) in disaster management: a conceptual model and its deployment
S. Afroogh, A. Mostafavi, A. Akbari, Y. Pouresmaeil, S. Goudarzi, F. Hajhosseini, K. Rasoulkhani
AI and Ethics, 2023
@article{afroogh2024embedded,
title = {Embedded ethics for responsible artificial intelligence systems ({EE-RAIS}) in disaster management: a conceptual model and its deployment},
author = {Afroogh, Saleh and Mostafavi, Ali and Akbari, Ali and Pouresmaeil, Yasser and Goudarzi, Sajedeh and Hajhosseini, Faegheh and Rasoulkhani, Kambiz},
journal = {AI and Ethics},
volume = {4},
number = {4},
pages = {1117--1141},
year = {2024},
doi = {10.1007/s43681-023-00309-1}
}
Tracing app technology: an ethical review in the COVID-19 era and directions for post-COVID-19
S. Afroogh, A. Esmalian, A. Mostafavi, et al.
Ethics and Information Technology, 2022
@article{afroogh2022tracing,
title = {Tracing app technology: an ethical review in the {COVID-19} era and directions for post-{COVID-19}},
author = {Afroogh, Saleh and Esmalian, Amir and Mostafavi, Ali and Akbari, Ali and Rasoulkhani, Kambiz and Esmaeili, Shahriar and Hajiramezanali, Ehsan},
journal = {Ethics and Information Technology},
volume = {24},
number = {3},
pages = {30},
year = {2022},
doi = {10.1007/s10676-022-09659-6}
}
A probabilistic theory of trust concerning artificial intelligence: can intelligent robots trust humans?
S. Afroogh
AI and Ethics, 2022
@article{afroogh2023probabilistic,
title = {A probabilistic theory of trust concerning artificial intelligence: can intelligent robots trust humans?},
author = {Afroogh, Saleh},
journal = {AI and Ethics},
volume = {3},
number = {2},
pages = {469--484},
year = {2023},
doi = {10.1007/s43681-022-00174-4}
}
The Mirage of Motivation Reason Internalism
S. Afroogh
The Journal of Value Inquiry, 2022
@article{afroogh2024mirage,
title = {The mirage of motivation reason internalism},
author = {Afroogh, Saleh},
journal = {The Journal of Value Inquiry},
volume = {58},
number = {1},
pages = {111--129},
year = {2024},
doi = {10.1007/s10790-021-09871-5}
}
COVID-19, Scarce Resources and Priority Ethics: Why Should Maximizers Be More Conservative?
S. Afroogh, A. Kazemi, A. Seyedkazemi
Ethics, Medicine and Public Health, 2021
@article{afroogh2021covid,
title = {{COVID-19}, scarce resources and priority ethics: why should maximizers be more conservative?},
author = {Afroogh, Saleh and Kazemi, A. and Seyedkazemi, A.},
journal = {Ethics, Medicine and Public Health},
volume = {18},
pages = {100698},
year = {2021},
doi = {10.1016/j.jemep.2021.100698}
}
Empathic design in engineering education and practice: an approach for achieving inclusive and effective community resilience
S. Afroogh, A. Esmalian, J. P. Donaldson, A. Mostafavi
Sustainability, 2021
@article{afroogh2021empathic,
title = {Empathic design in engineering education and practice: an approach for achieving inclusive and effective community resilience},
author = {Afroogh, Saleh and Esmalian, Amir and Donaldson, Jonan Phillip and Mostafavi, Ali},
journal = {Sustainability},
volume = {13},
number = {7},
pages = {4060},
year = {2021},
doi = {10.3390/su13074060}
}
Full list on Google Scholar
Datasets
Dataset · 2026
KidAID
Real-world AI incidents involving children and adolescents, coded by sector, harm category, failure mode and severity.
Harvard Dataverse →
Dataset · 2025
ChildSafeLLM
200 child-safety evaluation prompts for two age groups, with harmful-response flags and 0–5 action labels for eight LLMs.
Harvard Dataverse →
Dataset · 2024
AGGA
Academic guidelines for generative AI, collected from universities worldwide.
Harvard Dataverse →
Talks & Teaching
Talks
Sep 2026
“Introduction to AI and LLMs,” for undergraduate students, The University of Texas at Austin.
May 2024
“Large Language Models Ethics: Advancements, Challenges, and Future Directions.” 2nd Annual Smart Cities and AI Innovations Symposium.
Teaching
2021–2023
PHI 210 Symbolic Logic · State University of New York at Albany
2021
PHI 230 Morals and Society · State University of New York at Albany
2016–2018
PHIL 341 Symbolic Logic · Texas A&M University
2016
PHIL 482 / ENGR 482 Ethics and Engineering · Texas A&M University
Mentees
Abhejay MuraliSoftware Engineer · Charles Schwab
Kerem ErsözSoftware Engineer II · Microsoft
Austin DengTuring Scholar (CS Honors) · UT Austin
Taegang KimAI Researcher · Urban Information Lab, UT Austin
Shrestha MishraTechnical Lead · TANTV
Isha AgrawalCS + Math · UT Austin
Paintings
In my spare time I paint.
