Chinese AI Language Models Polluted by Inappropriate Internet Content, Study Finds

The article clearly involves AI systems (conversational AI like ChatGPT) influencing human behavior and language use. The reported changes are indirect effects of AI use, with potential long-term societal and cultural harms suggested but not yet realized. Since no direct or indirect harm has occurred yet, but there is a credible plausible risk of future harm (e.g., loss of linguistic diversity, impact on human cognition and social behavior), this fits the definition of an AI Hazard. The article does not describe an AI Incident because no actual harm has been caused or documented. It is not merely complementary information because the main focus is on the potential negative impact of AI on human language and behavior, not on responses or ecosystem updates. Therefore, the correct classification is AI Hazard.

The article explicitly involves AI systems (large language models) and their training data. It details how polluted data (containing adult and gambling content) is embedded in the models' token vocabularies, leading to hallucinations and unreliable outputs. This represents a malfunction or problematic use stemming from AI development and training data sourcing. However, the article does not report any specific realized harm such as injury, rights violations, or community harm caused by these AI outputs. Instead, it warns of the plausible future risks and systemic vulnerabilities arising from this data pollution. Hence, it fits the definition of an AI Hazard rather than an AI Incident. It is not Complementary Information because it is not an update or response to a previously known incident, nor is it unrelated as it clearly concerns AI system development and its risks.

The article explicitly involves AI systems (large language models like GPT-4o) and discusses their development and use, focusing on the data pollution problem. The presence of polluted tokens and their impact on AI outputs is a malfunction or deficiency in the AI system's training and behavior. While no direct harm (such as injury, rights violations, or property damage) is reported, the article clearly outlines how this pollution could plausibly lead to harms like misinformation, hallucinations, and degraded user trust or experience. The article does not describe a realized incident but warns of a credible risk inherent in the AI system's training data and behavior. Hence, it fits the definition of an AI Hazard rather than an AI Incident or Complementary Information. It is not unrelated because it centers on AI system issues and their implications.

The event involves AI systems (large language models) and their development process (training data contamination). Although no direct harm is reported as having occurred, the contamination causes AI hallucinations and unsafe outputs, which plausibly could lead to harms such as misinformation, exposure to harmful content, or violation of user rights. The article focuses on the potential risks and systemic vulnerabilities rather than a concrete realized harm. Hence, it fits the definition of an AI Hazard rather than an AI Incident or Complementary Information. It is not unrelated because it clearly involves AI systems and their impact.

The event involves an AI system (large language models like GPT-4o) whose development (training on polluted data) and use (generating hallucinated or inappropriate content) have directly led to harm. The harms include misinformation, degraded user experience, and potential violation of rights to accurate information, which fall under harm to communities and violation of rights. The article provides concrete evidence from research quantifying the extent of data pollution and its effects on AI outputs, demonstrating realized harm rather than just potential risk. Hence, it meets the criteria for an AI Incident rather than a hazard or complementary information.