Abuse & Misuse Accidents Accountability Accuracy Adversarial Prompts Agential AGI - AI technology itself Effects on humans and other living beings: Existential risks Non-existential removing from the control of human owners/managers AGIs being given or developing unsafe goals with poor ethics morals values death development Ethics Influence jurisprudence Law Regulation leads to losing future Race (Environmental/Structural) Risk Management Security Society Trust AI's inherent safety Alignment failures in existing ML systems Allocative Harms Amplification biases Anthropomorphism Appropriate Relationships Art Creativity Automation Access Environmental Beneficial Between-country issues: global inequality Bias fairness Broken By Mistake Post-Deployment Pre-Deployment Capabilities that increase likelihood existential risk Capability Causes Misalignment CBRN Information Challenges associated economy: Child sexual exploitation Competing for jobs Completeness Complexity Intended Task Usage Environment Compliance Confabulation Content Safety Risks Continual Learning Controllability Controversial Opinions Cooperation Copyright Authorship Corrigibility Culture identity Cyber-offense Cybercrime Dangerous capabilities Dangerous Violent Hateful Data Breach/Privacy Liberty Governance Issues Privacy Protection/Privacy Quality Data-level Deception Defamation Degree Automation Control Design Development Diluting Rights Direct catastrophe Discrimination Discrimination Exclusion Toxicity Disruption Concerns Domain-specific Double edge components Economic Education Elections Emergent functionality Energy Consumption Enfeeblement Societal Impact Impacts Eroded epistemics Erosion Ethical social Morality Evaluation Auditing Evolution Exacerbating Climate Change Market Power Concentration Explainability Reasoning Transparency Explainable Extintion Extreme Fairness First-Order Foregone benefits Goal-related Hallucinations Harassment Impersonation Extortion Hardware Vulnerabilities Harm caused by incompetent unaligned competent Harmful Homogenization responses Hate Healthcare Human Agency Oversight Autonomy Intregrity Human-AI Configuration interaction Human-Computer Interaction Human-like immoral decisions Human–AI Illegal Activities Intellectual Property Impacts: The Technical Base System Inadequate management Inclusive Growth Well-being Incompetence Independently Indirect contributions Indiscriminate weapons (CBRNE) Individual needs Inequality Hazards Integrity Manipulation Injustice Instruction Attacks property rights violations Intelligibility Intentional: socially accepted/legal condemned/illegal Interpersonal Interpretability External Tools Labor displacement impact Manipulation Theft Displacement Lack transparency interpretability Legal Rights-Related challenges Lethal Autonomous Weapons (LAW) Liability Long-horizon planning Long-term Loss oversight an autonomous model then taking harmful actions privacy Malicious Use (Intentional) Unleashing Agents use Uses Malign belief distributions Mental Health Meta-cognition Mind Misaligned Behaviors Misapplication a non-malicious actor Miscellaneous Misinformation Misleading Mobility Model Model-level Moral Multi-agent Negative impacts Non-violent crimes Not-Suitable-for-Work (NSFW) Obscene Degrading and/or Abusive Offensiveness On Purpose Post Deployment Opacity Opaque Collection Organizational (Accidental) Over-dependence Overreliance systems which cannot be subsequently unpicked Performance Robustness Personal Identity Theft Persuasion manipulation Physical Physicalistic decision-making Political strategy Poor performance used its intended purpose example leading biased Power-seeking behavior Predictability Leakage regulation security Problem Identification Formulation Products Property/legal Protection Providing new malicious Proxy misspecification Quality-of-Service Rationality Reduced Autonomy/Responsibility Regulations policy Reliability Repeatability / Reproducibility Representation Representational Resistance Responsability Responsibility area 1: speech 2: 3: 4: 5: 6: Socioeconomic harms through Anthropomorphic Assistant Injury accelerating scientific progress Malfunctions Unreliability Rogue AIs (Internal) Safe learning Trustworthiness Applications Second-Order Defense Self-proliferation Semantic Communication Sex-related Sexual Single point failure Situational awareness Social justice Norm concerns around reduce realisation potential consequences Socio-technical Infrastructural environmental Software Specialized Advice Storage (Memory) Structural Subagents Suicide self-harm Sustainability Operational Systemic operational Technological Maturity Technological Analytical Technology Tendencies explainability reliability Trustworthy Type Diffusion responsibility Bigger than expected Worse Willful indifference Criminal weaponization State Weaponization Typical scenarios Unauthorized Undesirable Cases Unfairness Unhelpful Unintended outcomes interactions Unintentional: direct indirect Untruthful Usurpation automation Value Chain Component Integration lock-in specification Verifiability acquisition Within-country domestic Worsened conflict epistemic processes society Writing Research