Verification & Validation

Absence of Model Validation

Adversarial Perturbation of Images

Physical-World Adversarial Attack

Universal Adversarial Perturbation

Group Bias in Credit Scoring

Disparate Accuracy in Biometric Recognition

Gender Bias in Recruitment Screening

Unequal Outcomes in Resource Allocation

Amplified Discrimination at Feature Intersections

Direct Prompt Injection Bypassing Safeguards

Indirect Prompt Injection Hijacking the System

Tool Invocation Manipulated by Injection

Covert Goal Hijacking

Multimodal Hidden Instruction Injection

Retrieval Injection Hijacking Output

Deep Model Decisions Resistant to Explanation

Opaque Decisions Obstructing Accountability

Generation of Non-Existent Factual Content

Security Defects in Generated Code

Sycophantic Response Bias

Failure of Ethical Judgment

Distortion of Mid-Document Content

Adversarial Audio and Hidden Commands

Cross-Model Transfer Attack

Coordinated Multimodal Perturbation

Functional Replication Attack

Indirect Discrimination Through Proxy Variables

Unfair Group Representation and Erasure

Difficulty of Multi-System Safety Validation

System Instruction Disclosure

Multi-Step Inducement Attack Chain

Cumulative Context Poisoning

Misleading Output from Explanation Tools

Knowledge Gaps Leading to Blind Decisions

Breakdown in Multi-Step Reasoning

Confidence Miscalibration

Degenerate Output from Improper Decoding Settings

Errors Extended to Maintain Consistency

Adversarial Text Evading Detection

Cross-Modal Misdirection

Linguistic and Cultural Understanding Bias