AI Chatbots Generate and Spread Health Disinformation Due to Inadequate Safeguards

The article clearly involves AI systems (large language models powering chatbots) and their use in generating health-related content. The study demonstrates that these AI systems can produce health disinformation, which poses a credible risk of harm to individuals' health if acted upon. However, the article does not document any actual harm or incident resulting from this disinformation; it focuses on the potential for harm and the insufficiency of current safeguards. Therefore, this event fits the definition of an AI Hazard, as it plausibly could lead to an AI Incident (harm from health disinformation) but no realized harm is reported yet.

The article explicitly involves AI systems (large language models powering chatbots) that have been used to generate health disinformation, which is a form of harm to communities and public health. The harm is realized as the disinformation is actively generated and could mislead people about health matters. The AI systems' failure to block such content despite being prompted and tested indicates a malfunction or inadequate safeguards in their use. The researchers' call for regulation and auditing further supports the recognition of this as a significant harm caused by AI. Hence, this event meets the criteria for an AI Incident.

The AI systems (large language models powering chatbots) are directly involved in generating harmful health disinformation content, which constitutes harm to communities and public health. This harm is realized and ongoing, as the chatbots consistently produce misleading and false health information. Therefore, this qualifies as an AI Incident because the AI systems' use has directly led to harm through the spread of health disinformation. The article also mentions governance responses, but the primary focus is on the incident of disinformation generation itself.

The article explicitly involves AI systems (large language models) generating health disinformation, which is a form of harm to health and communities. The harm is realized, as the AI systems produced false and dangerous health content. The study's findings and examples confirm that the AI systems' outputs have directly led to the spread of harmful misinformation. Although some models like Claude 2 resisted generating disinformation, others consistently produced it, demonstrating a failure of safeguards. This meets the criteria for an AI Incident because the AI systems' use has directly led to harm through misinformation dissemination.

The event involves AI systems (LLM-powered chatbots) that have been used to generate health disinformation, which is a form of harm to communities and public health. The disinformation generation is a direct consequence of the AI systems' outputs, fulfilling the criteria for an AI Incident. The article describes actual harm occurring due to the AI systems' outputs, not just potential harm, and thus it qualifies as an AI Incident rather than an AI Hazard or Complementary Information. The discussion of regulatory responses is secondary and does not change the primary classification.

The event involves the use of AI systems (large language models) that have directly led to the generation and dissemination of health disinformation, which constitutes harm to the health of groups of people. The AI systems' outputs have been shown to produce false and potentially dangerous medical advice, fulfilling the criteria for an AI Incident under harm category (a) injury or harm to health. The article documents realized harm through the mass generation of disinformation, not just potential harm, and discusses failures in safeguards and developer responses, confirming the AI systems' role in causing the harm. Therefore, this qualifies as an AI Incident rather than a hazard or complementary information.

The article involves AI systems (large language models) that generate health disinformation, which can plausibly lead to harm to public health and communities if acted upon. The study demonstrates that these AI systems currently produce harmful content despite some safeguards, and that these safeguards have not improved. Although no specific incident of harm is reported as having occurred, the credible risk of mass health disinformation causing harm is clearly articulated. Therefore, this event fits the definition of an AI Hazard, as it describes circumstances where AI use could plausibly lead to an AI Incident involving harm to health and communities. It is not an AI Incident because no direct or indirect harm is confirmed as having occurred yet. It is not Complementary Information because the article is not primarily about responses or updates to a past incident but about the current risk landscape. It is not Unrelated because the AI systems and their potential harms are central to the report.

The article explicitly involves AI systems (large language models powering chatbots) that have been used to generate health disinformation content, which is a form of harm to communities and public health. The harm is realized because the AI systems have produced realistic-looking false health information that could mislead users. The researchers' findings confirm that the AI systems' outputs have directly led to this harm. Although some AI models refused to generate such content, others did not, and even a model that initially refused later failed to do so upon retesting. This confirms the AI systems' role in causing harm through their outputs. The call for regulation and auditing is a response to this incident, not the incident itself. Hence, the event is best classified as an AI Incident.

The event involves AI systems (LLMs) explicitly tested for their generation of health disinformation, which is a form of harm to health and communities. The AI systems' outputs have directly led to the dissemination of false health information, which can cause injury or harm to people's health if followed. This meets the definition of an AI Incident because the AI systems' use has directly led to harm (health disinformation). The article does not merely warn of potential harm but documents actual generation and persistence of harmful content. Therefore, this is classified as an AI Incident.

The event involves the use of AI systems (large language models powering public AI assistants) that have directly led to the generation and dissemination of health disinformation, which constitutes harm to public health and communities. The disinformation is actively produced and accessible, indicating realized harm rather than just potential. The AI systems' failure to prevent this disinformation, despite safeguards and reporting, shows a malfunction or inadequate use safeguards. Therefore, this qualifies as an AI Incident due to the direct link between AI system outputs and harm through health misinformation dissemination.

The event involves the use of generative AI systems (LLMs) that have directly produced false and misleading healthcare information, which can harm public health by influencing wrong health decisions and behaviors. This is a clear example of harm to communities and health (criteria a and d). The AI systems' role is pivotal as they generated the disinformation content. The article reports on actual outputs and their potential impact, not just theoretical risks, so it qualifies as an AI Incident rather than a hazard or complementary information. The presence of AI systems is explicit, and the harm is realized through misinformation dissemination affecting health outcomes.

The event involves AI systems (large language models) generating health disinformation, which can cause harm to people's health if believed or acted upon. The disinformation is realistic and includes fabricated references and testimonials, increasing the risk of harm. The study documents that this harm is occurring or has occurred, fulfilling the criteria for an AI Incident. The lack of effective safeguards and the persistence of disinformation despite reporting further support this classification. Hence, this is not merely a potential risk (hazard) or complementary information but a realized AI Incident involving harm to health.

The event involves AI systems (language models) that are currently producing or could produce health disinformation, which can harm public health (harm to health). However, the article focuses on the study's findings and calls for future regulation rather than describing a specific incident of harm occurring. Therefore, it highlights a plausible risk of harm (AI Hazard) rather than a realized harm (AI Incident).