This paper outlines the ethical and social risks associated with large-scale language models (LMs), aiming to structure the risk landscape for responsible innovation. It identifies 21 risks across six categories: Discrimination, Exclusion and Toxicity; Information Hazards; Misinformation Harms; Malicious Uses; Human-Computer Interaction Harms; and Automation, Access, and Environmental Harms. The first category includes risks such as unfair discrimination, exclusionary norms, toxic language, and lower performance for some social groups, stemming from training data that includes harmful language and overrepresentation of certain identities. The second category involves risks from private data leaks or correct inference of private information. The third category covers risks from false or misleading information, which can lead to material harm or unethical actions. The fourth category includes risks from users or developers using LMs to cause harm, such as spreading disinformation or creating scams. The fifth category focuses on risks from conversational agents that interact directly with users, including overestimation of capabilities and privacy violations. The sixth category includes environmental, access, and inequality risks from LM use. The paper discusses the origins of these risks and potential mitigation strategies, emphasizing the need for a broad, inclusive approach to address ethical and social risks. It highlights the importance of collaboration, research, and policy in mitigating these risks and ensuring responsible innovation. The report also notes the limitations of current research, including the need for further study on long-term risks and multi-modal models.This paper outlines the ethical and social risks associated with large-scale language models (LMs), aiming to structure the risk landscape for responsible innovation. It identifies 21 risks across six categories: Discrimination, Exclusion and Toxicity; Information Hazards; Misinformation Harms; Malicious Uses; Human-Computer Interaction Harms; and Automation, Access, and Environmental Harms. The first category includes risks such as unfair discrimination, exclusionary norms, toxic language, and lower performance for some social groups, stemming from training data that includes harmful language and overrepresentation of certain identities. The second category involves risks from private data leaks or correct inference of private information. The third category covers risks from false or misleading information, which can lead to material harm or unethical actions. The fourth category includes risks from users or developers using LMs to cause harm, such as spreading disinformation or creating scams. The fifth category focuses on risks from conversational agents that interact directly with users, including overestimation of capabilities and privacy violations. The sixth category includes environmental, access, and inequality risks from LM use. The paper discusses the origins of these risks and potential mitigation strategies, emphasizing the need for a broad, inclusive approach to address ethical and social risks. It highlights the importance of collaboration, research, and policy in mitigating these risks and ensuring responsible innovation. The report also notes the limitations of current research, including the need for further study on long-term risks and multi-modal models.