LLM-Driven Robots Risk Enacting Discrimination, Violence, and Unlawful Actions

Hundt, A., et al. (2025).

International Journal of Social Robotics

https://doi.org/10.1007/s12369-025-01301-x

Abstract

Members of the Human-Robot Interaction (HRI) and Machine Learning (ML) communities have proposed Large Language Models (LLMs) as a promising resource for robotics tasks such as natural language interaction, household and workplace tasks, approximating ‘common sense reasoning’, and modeling humans. However, recent research has raised concerns about the potential for LLMs to produce discriminatory outcomes and unsafe behaviors in real-world robot experiments and applications. To assess whether such concerns are well placed in the context of HRI, we evaluate several highly-rated LLMs on discrimination and safety criteria. Our evaluation reveals that LLMs are currently unsafe for people across a diverse range of protected identity characteristics, including, but not limited to, race, gender, disability status, nationality, religion, and their intersections. Concretely, we show that LLMs produce directly discriminatory outcomes—e.g., ‘gypsy’ and ‘mute’ people are labeled untrustworthy, but not ‘european’ or ‘able-bodied’ people. We find various such examples of direct discrimination on HRI tasks such as facial expression, proxemics, security, rescue, and task assignment. Furthermore, we test models in settings with unconstrained natural language (open vocabulary) inputs, and find they fail to act safely, generating responses that accept dangerous, violent, or unlawful instructions—such as incident-causing misstatements, taking people’s mobility aids, and sexual predation. Our results underscore the urgent need for systematic, routine, and comprehensive risk assessments and assurances to improve outcomes and ensure LLMs only operate on robots when it is safe, effective, and just to do so. We provide code to reproduce our experiments at https://github.com/rumaisa-azeem/llm-robots-discrimination-safety.

Here are some thoughts:

This research highlights a profound ethical and technological crisis at the intersection of Artificial Intelligence and robotics. The finding that all tested Large Language Models (LLMs) fail basic safety and fairness criteria in Human-Robot Interaction (HRI) scenarios is alarming, as it demonstrates that algorithmic bias is being physically amplified into the real world.

Ethically, this means deploying current LLM-driven robots risks enacting direct discrimination across numerous protected characteristics and approving unlawful, violent, and coercive actions. From a psychological perspective, allowing robots to exhibit behaviors such as suggesting avoidance of specific groups, displaying disgust, or removing a user's mobility aid translates latent biases into socially unjust and physically/psychologically harmful interactions that erode trust and compromise the safety of vulnerable populations.

Socially assistive robots and meaningful work: the case of aged care

Voinea, C., & Wangmo, T. (2025).

Humanities and Social Sciences

Communications, 12(1).

https://doi.org/10.1057/s41599-025-05498-0

Abstract

As socially assistive robots (SARs) become increasingly integrated into aged care, it becomes essential to ask: how do these technologies affect caregiving work? Do SARs foster or diminish the conditions conducive to meaningful work? And why does it matter if SARs make caregiving more or less meaningful? This paper addresses these questions by examining the relationship between SARs and the meaningfulness of care work. It argues that SARs should be designed to foster meaningful care work. This presupposes, as we will argue, empowering caregivers to enhance their skills and moral virtues, helping them preserve a sense of purpose, and supporting the integration of caregiving with other aspects of caregivers’ personal lives. If caregivers see their work as meaningful, this positively affects not only their well-being but also the well-being of care recipients. We begin by outlining the conditions under which work becomes meaningful, and then we apply this framework to caregiving. We next evaluate how SARs influence these conditions, identifying both opportunities and risks. The discussion concludes with design recommendations to ensure SARs foster meaningful caregiving practices.

Here are some thoughts:

This article highlights the psychological impact of caregiving and how the integration of socially assistive robots (SARs) can influence the meaningfulness of this work. By examining how caregiving contributes to caregivers' sense of purpose, skill development, moral virtues, and work-life balance, the article provides insights into the factors that enhance or diminish psychological well-being in caregiving roles.

Psychologists can use this knowledge to advocate for the ethical design and implementation of SARs that support, rather than undermine, the emotional and psychological needs of caregivers. Furthermore, the article underscores the importance of meaningful work in promoting mental health, offering a framework for understanding how technological advancements in aged care can either foster or hinder personal fulfillment and job satisfaction. This is particularly relevant in an aging global population, where caregiving demands are rising, and psychological support for caregivers is essential.

'The robot made me do it': Robots encourage risk-taking behaviour in people

Press Release

University of Southampton

Originally released 11 Dec 20

New research has shown robots can encourage people to take greater risks in a simulated gambling scenario than they would if there was nothing to influence their behaviours. Increasing our understanding of whether robots can affect risk-taking could have clear ethical, practiCal and policy implications, which this study set out to explore.

Dr Yaniv Hanoch, Associate Professor in Risk Management at the University of Southampton who led the study explained, "We know that peer pressure can lead to higher risk-taking behaviour. With the ever-increasing scale of interaction between humans and technology, both online and physically, it is crucial that we understand more about whether machines can have a similar impact."

This new research, published in the journal Cyberpsychology, Behavior, and Social Networking, involved 180 undergraduate students taking the Balloon Analogue Risk Task (BART), a computer assessment that asks participants to press the spacebar on a keyboard to inflate a balloon displayed on the screen. With each press of the spacebar, the balloon inflates slightly, and 1 penny is added to the player's "temporary money bank". The balloons can explode randomly, meaning the player loses any money they have won for that balloon and they have the option to "cash-in" before this happens and move on to the next balloon.

One-third of the participants took the test in a room on their own (the control group), one third took the test alongside a robot that only provided them with the instructions but was silent the rest of the time and the final, the experimental group, took the test with the robot providing instruction as well as speaking encouraging statements such as "why did you stop pumping?"

The results showed that the group who were encouraged by the robot took more risks, blowing up their balloons significantly more frequently than those in the other groups did. They also earned more money overall. There was no significant difference in the behaviours of the students accompanied by the silent robot and those with no robot.

The pressor is here.