Consumer Reports: AI Chatbots Give Risky Health and Safety Advice
Experiments with five general-purpose AI bots reveal dangerous gaps in safety recommendations, from carbon monoxide alarms to water filtration.
General-purpose AI chatbots frequently provide incomplete or dangerous health and safety advice, according to a study by Consumer Reports. The research warns that relying on these tools for critical safety information can lead to physical harm.
Researchers tested five different AI chatbots to determine if their responses aligned with expert recommendations. The results showed significant failures across multiple platforms. Google Gemini, for instance, incorrectly suggested that users place a carbon monoxide alarm inside an attached garage. Expert guidance, including that from Consumer Reports, advises against this because a lack of temperature control in garages can damage the alarm's sensor or battery.
Other bots struggled with technical safety specifications. Microsoft Copilot provided insufficient details regarding PFAS water filtration, failing to emphasize the necessity of NSF/ANSI 53 certification to ensure the filter is effective. Similarly, Meta AI failed to highlight the severe safety risks associated with water beads, offering only a passing reference to parental supervision rather than detailing the potential for serious injury.
The Risk of AI Reliance
This trend comes as tools like ChatGPT, Gemini, and Copilot become deeply integrated into daily routines. Many users now turn to these large language models (LLMs) for immediate answers to health questions. However, because these are general-purpose tools rather than specialized medical or safety systems, they are prone to "hallucinations" or oversimplifying complex safety protocols.
When an AI provides a confident but incorrect answer about a safety device or a toxic substance, the consequences can be life-threatening. The gap between a "helpful" AI response and a verified expert recommendation can be the difference between a functioning safety alarm and a failed one during an emergency.
Moving Toward Verification
The study underscores a critical need for users to treat AI as a starting point rather than a final authority. Consumer Reports suggests that users should view AI as an assistant rather than an expert, recommending that people ask questions multiple times to compare answers and cross-reference them with primary sources.
As AI continues to evolve, the industry faces pressure to improve the accuracy of safety-critical information. For now, the burden of verification remains with the user. Experts advise that any health or safety recommendation generated by an AI must be verified through a certified professional or an official safety organization before being implemented in the home.