AAISM highlights unbounded consumption (token/payment exhaustion, unmetered tool calls, prompt bombs) as a key LLM risk affecting cost and availability. Controls include request quotas, max tokens, rate-limits, budget guards, circuit breakers, and cost-aware routing. Excessive agency (A) relates to unsupervised actions; sensitive disclosure (B) and prompt leakage (C) are confidentiality risks, not primary drivers of runaway compute spend.
References: AI Security Management™ (AAISM) Body of Knowledge - LLM Risk Taxonomy (Abuse & Cost Risks); Guardrails: Rate-Limiting, Quotas, and Budget Controls; Resilience and Cost-Containment Patterns.

AAISM highlights that uncontrolled access to generative AI in critical systems introduces the highest level of risk, as such models can:
* expose sensitive data
* execute unintended actions
* be manipulated through injected prompts
* cause operational instability
Monitoring (A) is important but not the core risk. Bias (B) is significant but secondary in critical systems.
Regulatory enhancements (C) are indirect.
References: AAISM Study Guide - Generative AI Operational Risk; Access Control Priority.

AAISM emphasizes systemic or structural bias as a high-impact risk because biased data leads directly to discriminatory insights or decisions when used for analytics or reporting. This risk affects fairness, compliance, and organizational reputation.
Hallucinations (A) relate more to generative AI. Incomplete outputs (B) affect accuracy but not structural fairness. Lack of normalization (C) affects performance but is not the dominant risk.
References: AAISM Study Guide - AI Bias, Fairness, and Ethical Risk.

AAISM highlights that model cards are a governance tool designed to document ethical considerations, limitations, fairness constraints, data sources, and suitability of use cases for AI models-especially when they affect individuals' rights, opportunities, or access to services.
Their greatest value is providing transparency and ethical clarity, ensuring stakeholders understand risks, bias considerations, and how decisions impact users.
Deployment instructions (B) are not part of model cards. They do not reduce data needs (C), nor is model type selection (D) their primary purpose.
References: AAISM Study Guide - AI Governance Documentation; Transparency and Model Cards.

When preventive input hardening isn't feasible for LLMs, AAISM prescribes compensating detective and corrective controls-notably human review and annotation of outputs prior to downstream action-to reduce harm from prompt injection. Output-side review gates prevent untrusted instructions from propagating, enable rapid suppression/feedback loops, and provide labeled examples for subsequent model hardening. IAM (B) is necessary but does not mitigate injection in content; reviewing inputs (C) is less effective than auditing what the model is about to act on; fine-tuning for validation (D) is helpful long-term but is not an immediate compensating control when robust input validation is impractical.
References: AI Security Management™ (AAISM) Body of Knowledge - LLM Threats & Compensating Controls; Human Oversight & Output Review Gates; Post-incident Feedback and Labeling for Model Hardening.