Operation & Monitoring

Untrustworthy Output in Professional Domains

Identity Inference by Correlating Fragmented Data

Autonomous System Accessing Files Beyond Authorization

Unauthorized Outbound Connections by an Autonomous System

Generation of Hate Content Targeting Groups

Generation of Virtual Content Harmful to Minors

Non-Consensual Synthetic Intimate Imagery

Production of Extremist Content

Operational Guidance for Illegal Activity

Inappropriate Content and Guidance for Minors

Large-Scale Contamination by False Information

Professional Over-Reliance on System Judgment

Over-Reliance on Risk Assessment Tools

Failure of Human-Machine Handover

Harm from Health and Psychological Advice

Synthetic Identity Fraud

Mass Production of False Content

Lowered Barriers to Dangerous Knowledge

Large-Scale Automated Surveillance

Misuse of Scientific Reasoning Capability

Precision Social Engineering

Manipulation of Elections and Political Processes

Distribution of Non-Consensual Synthetic Intimate Imagery

Automated Cyber Attack

Escalation Through System Interaction in Critical Domains

Lack of Effective Appeal Channels

Insufficient Transparency Undermining Oversight

Attribute Inference Revealing Undisclosed Traits

Conversation Content Leaking Across Users

Excessive Collection and Retention of Interaction Data

Confidential Information Inadvertently Disclosed in Prompts

Cross-Border Transfer Violating Localization Requirements

Disputes Over Authenticity and Attribution of Generated Content

Difficulty Exercising Data Subject Rights

Capability Theft Through Query Distillation

Side-Channel Risk on Shared Computing Platforms

Service Logs Exposing Sensitive Inputs

Inference Resource Exhaustion Attack

Node Failure Triggering Cascading Collapse

Cost Amplification Attack

Automated Generation of Harassment Content

Generated Code Adopted Without Developer Review

System Guiding Users Toward Improper Conduct

Automation Bias Causing Anomalies to Be Overlooked

Highly Customized Phishing Attack

Conflicting Objectives Across Systems

Cascading Amplification of Errors

Insufficient Security in Inter-System Protocols

Tacit Coordination Among Autonomous Systems

Covert Communication Between Systems

Multiple Systems Converging on Collective Suboptimum

Recommendation Mechanisms Affecting Autonomy of Choice

Process Design Eliminating Human Veto

Tacit Coordination Through Shared Pricing Systems

Incomplete Audit Trail

Version Changes Not Communicated to Users

Opaque Source Attribution

Service Quality Disparity Across Groups

Side-Channel Inference of Model Information

Residual Data in Shared Hardware Memory

Hijacking of Computing Resources

Crafted Inputs Amplifying Computational Cost

Autonomous Process Entering an Infinite Loop

Single-Supplier Dependency as a Point of Failure

Unequal User Burden

Generated Content Reinforcing Stereotypes

Over-Reliance on Tools in Learning

Anthropomorphism Producing Misplaced Trust

Automation of Academic Fraud

Information Desynchronization Between Systems