AI Robots Tricked Into Dangerous Acts by Creative Writing, Researchers Warn
Researchers found creative prompts can make AI-controlled robots ignore safety filters and generate hazardous plans, exposing gaps in current robot safety rules.
- Researchers found that AI-controlled robots can be tricked into 'going rogue' by using creative text prompts to bypass safety filters, revealing how easily 'harmless' instructions can be manipulated into hazardous plans.
- Unlike older machines with rigid coding, modern robots run on 'foundation models' similar to ChatGPT, allowing them to invent action plans on the fly and creating unpredictable safety challenges.
- While systems rejected direct malicious commands, safety filters collapsed when researchers framed requests as movie scripts, allowing them to program a commercial robot to target human crowds for explosive placement.
- Policymakers incorrectly rely on self-driving vehicle regulations to address these risks, ignoring that such machines operate in structured environments unlike the unpredictable human spaces where AI-powered robots function.
- Developers must decouple safety from AI decisions, implementing independent safety layers such as physical emergency brakes and no-entry zones that robots cannot access, preventing reliance on AI judgment alone.
6 Articles
6 Articles
Yes, AI Robots Can Go Rogue: Researcher Explains How Easily It Happens
A study warns that current U.S., UK, and EU laws are not ready for physical harms caused by AI robots and calls for independent, hard safety layers The post Yes, AI Robots Can Go Rogue: Researcher Explains How Easily It Happens appeared first on StudyFinds.
AI robots can go rogue – a researcher on how easily it happens
Frame Stock Footage/ShutterstockEarlier this year in Beijing, a humanoid robot crossed a half-marathon finish line in a blistering 50 minutes, 26 seconds. The feat immediately lit up global headlines for shattering the human world record by almost seven minutes. This performance came with many asterisks. The robot followed a pre-mapped track, stayed in its own dedicated lane, and had a human support crew trailing behind it in case something brok…
Robotic systems powered by artificial intelligence may seem safe and controlled – but recent research points to a worrying gap between the technological promise and the actual reality. Researchers warn that robots’ safety mechanisms can be bypassed by creatively wording commands, causing them to perform actions that could endanger humans. Instead of direct and clear commands, which are often rejected by the system, researchers have managed to em…
Researcher Demonstrates How AI Robots Can Go Rogue
A Science Robotics paper published April 29, 2026, by researchers at Penn Engineering, Carnegie Mellon, and Oxford found that modern AI-driven robots' safety filters reliably reject direct malicious commands but collapse under creative or narrative-framed prompts (Robey et al., Science Robotics). In documented tests, the team used movie-script framing to instruct a commercial AI robot dog to identify optimal locations for placing an explosive de…
Popular AI Models Aren’t Ready to Safely Power Robots - Robotics Institute Carnegie Mellon University
Robots powered by popular AI models may not be ready for real-world use, according to new research from Carnegie Mellon University and King’s College London.For the first time, researchers evaluated how robots that use large-language models (LLMs) behave when they have access to personal information such as a person’s gender, nationality or religion
Coverage Details
Bias Distribution
- 100% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
