Continuous Monitoring of the AI Threat Landscape
5 minutes
5 Questions
Continuous Monitoring of the AI Threat Landscape is the ongoing, systematic process of observing, collecting, and analyzing information about emerging threats, vulnerabilities, adversarial techniques, and regulatory shifts that affect AI systems throughout their lifecycle. Unlike point-in-time assessments, it treats AI security as a dynamic discipline, recognizing that models, data pipelines, and attacker capabilities evolve rapidly. In advanced AI Security Management, this practice spans several layers. At the technical layer, organizations monitor model behavior for signs of adversarial inputs, prompt injection, data poisoning, model extraction, membership inference, and drift that may indicate manipulation or degradation. Telemetry from inference endpoints, access logs, and anomaly detection systems feeds into security operations workflows, often integrated with SIEM and SOAR platforms. At the intelligence layer, teams track threat feeds, academic research, vulnerability disclosures (such as those in MITRE ATLAS or the OWASP Top 10 for LLMs), and dark-web chatter about new attack methods targeting machine learning frameworks, open-source models, or supply-chain dependencies like pretrained weights and datasets. From an AI Risk Management perspective, continuous monitoring supports frameworks such as the NIST AI RMF (particularly the 'Measure' and 'Manage' functions) and ISO/IEC 42001 by ensuring that risk registers remain current. Identified threats are mapped to assets, likelihood and impact are reassessed, and controls are updated accordingly. This closed-loop approach enables rapid patching, model retraining, policy adjustment, and red-team re-engagement when new attack vectors appear. Mature programs establish clear metrics (mean time to detect AI-specific incidents, coverage of monitored models, frequency of threat intelligence updates), define escalation paths, and assign ownership across security, data science, legal, and compliance teams. They also monitor the external environment, including evolving regulations like the EU AI Act, which can change risk tolerance and obligations. Ultimately, continuous monitoring transforms AI security from reactive incident response into proactive resilience, allowing organizations to anticipate attacker innovation and maintain trustworthy, compliant AI operations.
Continuous Monitoring of the AI Threat Landscape: A Complete AAISM Study Guide
Overview
This guide covers the AAISM (ISACA Advanced in AI Security Management) concept of Continuous Monitoring of the AI Threat Landscape, which sits within the AI Risk Management for Security domain. You will learn why this topic matters, what it actually is, how a mature program operates, and how to recognize and answer exam questions on it.
1. Why Continuous Monitoring of the AI Threat Landscape Is Important
Traditional security risk assessments were often point-in-time exercises: assess annually, document findings, and revisit next year. That model fails for AI for several reasons:
- The attack surface evolves faster than any review cycle. New jailbreak techniques, prompt injection variants, data poisoning methods, model extraction attacks, and adversarial example strategies are published weekly, often in academic preprints before any vendor advisory exists.
- AI systems are probabilistic and opaque. A control that mitigated a threat yesterday may be bypassed tomorrow by a slightly reworded prompt or a new encoding trick. You cannot assume a static control baseline remains effective.
- Dependencies change silently. Foundation models are updated by providers, open-source model weights are replaced, third-party plugins and agents are added, and ML frameworks (PyTorch, TensorFlow, LangChain, etc.) receive security patches. Each change can introduce new vulnerabilities.
- Regulatory and legal exposure is shifting. The EU AI Act, emerging U.S. state laws, sector regulators, and standards bodies (NIST, ISO/IEC 42001) continue to issue new expectations that redefine what counts as an unacceptable risk.
- Adversaries use AI too. Attackers leverage generative AI for phishing, deepfakes, automated vulnerability discovery, and malware generation, meaning the threat environment for all systems, not just AI systems, is changing.
From an ISACA perspective, the key principle is that risk is dynamic. A risk register that is not continuously refreshed with current AI threat intelligence quickly becomes misleading, leading to misallocated controls, false assurance to senior management, and uninformed risk acceptance decisions.
2. What Continuous Monitoring of the AI Threat Landscape Is
Continuous monitoring of the AI threat landscape is the ongoing, structured process of collecting, analyzing, prioritizing, and acting upon information about emerging threats, vulnerabilities, attack techniques, incidents, and regulatory developments relevant to an organization's AI systems, so that risk assessments, threat models, and controls remain current.
It is important to distinguish it from related concepts you will also see on the exam:
- Model performance monitoring (drift detection, accuracy tracking, bias metrics) looks inward at how a deployed model behaves. Threat landscape monitoring looks outward at what adversaries and the environment are doing.
- Security event monitoring (SIEM, SOC alerts) detects attacks currently happening against your infrastructure. Threat landscape monitoring identifies threats that may affect you, often before they are weaponized against you.
- Periodic risk assessment is a scheduled snapshot. Threat landscape monitoring is the continuous feed that keeps the snapshot from going stale between assessments and triggers out-of-cycle reassessments.
Together, these form a layered assurance picture. The exam expects you to understand that threat landscape monitoring is primarily a risk management and governance activity that informs technical teams, not merely a technical tooling exercise.
3. How It Works: Components of an Effective Program
3.1 Define scope and intelligence requirements
Start with the organization's AI inventory and use cases. A company running an internal LLM-based assistant has different intelligence needs than one deploying computer vision in autonomous equipment. Define priority intelligence requirements (PIRs): what threats, to which assets, with what business impact, do we most need to know about? Without this, monitoring produces noise rather than insight.
3.2 Identify and curate sources
Mature programs draw from a blend of sources, for example:
- Threat taxonomies and frameworks: MITRE ATLAS (Adversarial Threat Landscape for AI Systems), OWASP Top 10 for LLM Applications, OWASP Machine Learning Security Top 10, NIST AI RMF and the NIST adversarial ML taxonomy (AI 100-2).
- Vulnerability and advisory feeds: CVE/NVD entries for ML frameworks and libraries, model provider security bulletins, cloud provider AI service advisories, software bill of materials (SBOM) and emerging AI BOM data.
- Incident repositories: the AI Incident Database, AIAAIC repository, and vendor post-mortems that reveal real-world failure modes.
- Research and community: academic preprints, red-team disclosures, security conference talks, bug bounty reports, and practitioner communities.
- Information sharing bodies: sector ISACs/ISAOs, government cyber agencies, and AI-specific sharing initiatives.
- Regulatory and standards trackers: legislative updates, regulator guidance, and standards revisions that change compliance obligations.
- Internal telemetry: findings from your own red teaming, guardrail bypass logs, abuse reports, and model monitoring anomalies, which are often the earliest indicator that an external technique is being used against you.
3.3 Collect and normalize
Automate where possible: subscribe to feeds, use threat intelligence platforms (TIPs) that can ingest AI-specific indicators, and map incoming items to a common taxonomy such as MITRE ATLAS techniques. Normalization allows comparison and trending over time.
3.4 Analyze and triage for relevance
Not every published attack matters to every organization. Analysts assess each item against the AI inventory: Does it affect a model, framework, or provider we use? Does it target a use case we have deployed? Is it theoretical, proof-of-concept, or actively exploited? What is the likely impact and ease of exploitation? This produces a prioritized, contextualized view rather than a raw list.
3.5 Integrate into risk management
This is the step the AAISM exam emphasizes most. Validated intelligence must flow into:
- The AI risk register, updating likelihood, impact, and velocity ratings and adding new risk scenarios.
- Threat models for individual AI systems, so that design-stage assumptions are revisited.
- Control effectiveness assessments, identifying where guardrails, input validation, access controls, or monitoring rules need updating.
- Key risk indicators (KRIs) and risk appetite reviews, so leadership can see whether exposure is trending toward tolerance thresholds.
- Third-party risk management, when the intelligence concerns a model provider, data supplier, or plugin vendor.
3.6 Act and communicate
Define escalation paths and response playbooks: who is notified when a critical AI vulnerability emerges, what interim mitigations are applied, and when an out-of-cycle risk assessment or emergency change is triggered. Communicate in business terms to risk committees and the board, and in technical terms to engineering and SOC teams.
3.7 Measure and improve
Track metrics such as time from public disclosure to internal assessment, percentage of AI assets mapped to current threat taxonomies, number of risk register updates driven by intelligence, and control changes implemented as a result. Review the program itself periodically to ensure sources and PIRs remain aligned with the evolving AI portfolio.
4. Governance Considerations
- Ownership: Typically the CISO or AI risk function owns the process, with contributions from threat intelligence, data science, legal/compliance, and procurement. The exam often tests who should own or approve decisions; the answer is usually management with accountability for risk, informed by specialists.
- Integration with ERM: AI threat intelligence should feed the enterprise risk management framework rather than operate as a silo.
- Documentation: Maintain an audit trail showing how intelligence changed risk ratings and controls. This demonstrates due care to regulators and auditors.
- Three lines model: First line (AI product and engineering teams) applies mitigations, second line (risk and security management) runs and interprets monitoring, third line (internal audit) provides assurance that the process works.
5. How to Answer Exam Questions on This Topic
AAISM questions follow the classic ISACA style: scenario-based, four options, often with two plausible choices where you must pick the best, most important, primary, or first action. Apply these reasoning patterns:
- Favor continuous over periodic. If an option proposes an annual or quarterly review and another proposes ongoing monitoring integrated into risk processes, the continuous option is almost always correct for this concept.
- Favor risk integration over raw collection. Collecting feeds is necessary but not sufficient. The best answer usually involves updating the risk register, threat models, or controls based on the intelligence.
- Favor relevance and prioritization over volume. ISACA values focusing resources on threats that affect the organization's actual AI assets and use cases. An option that says monitor everything equally is a distractor.
- Favor governance and process over tools. If a question asks what is MOST important to establish, the answer is more likely a defined process with ownership and escalation than purchasing a specific platform.
- Favor management accountability. Risk acceptance, appetite changes, and prioritization are management decisions informed by, not delegated to, technical staff or vendors.
- Watch for the FIRST step. When asked what to do first, it is usually to understand the AI inventory and define intelligence requirements, or to assess the relevance of a new threat to the organization, before implementing controls.
- Recognize the outward-looking nature. If a question contrasts drift monitoring or SIEM alerts with threat landscape monitoring, remember that threat landscape monitoring addresses external emerging threats and environmental change.
6. Exam Tips: Answering Questions on Continuous Monitoring of the AI Threat Landscape
Tip 1: Look for the trigger phrase. Scenarios mentioning a newly published jailbreak, a vulnerability in an ML library, a vendor model update, or a new regulation are signaling this concept. The correct response typically involves assessing relevance and updating the risk assessment, not immediately shutting systems down or ignoring the news until the next review.
Tip 2: Eliminate point-in-time answers. Any option implying that a completed risk assessment remains valid until the next scheduled cycle contradicts the dynamic-risk principle and is almost certainly wrong.
Tip 3: Eliminate reactive-only answers. Options that rely solely on responding to incidents after they occur miss the proactive purpose of threat landscape monitoring.
Tip 4: Know the named frameworks. Be able to recognize MITRE ATLAS as the AI-specific adversarial technique knowledge base, OWASP Top 10 for LLM Applications as the common LLM vulnerability list, and NIST AI RMF as the overarching risk framework. Questions may use these as the correct structured source for mapping threats.
Tip 5: Connect to the risk register. When unsure, the option that ties monitoring output to the risk register, KRIs, or control updates is usually the strongest, because it demonstrates the intelligence is being acted upon within governance.
Tip 6: Think about the whole supply chain. The AI threat landscape includes model providers, training data sources, open-source components, plugins, and agents. An option that limits monitoring to in-house models alone is incomplete.
Tip 7: Remember communication. Intelligence that is not communicated to decision-makers and implementers has no value. Options emphasizing timely, audience-appropriate reporting to risk committees and engineering teams reflect mature practice.
Tip 8: Distinguish the roles. Internal audit provides assurance over the process; it does not run the monitoring. Data scientists supply technical interpretation; they do not set risk appetite. The CISO or risk owner coordinates and escalates. Misattributed roles are common distractors.
Tip 9: Apply the business lens. ISACA exams reward answers that translate technical threats into business impact and align responses with risk appetite and organizational objectives.
Tip 10: Practice the reasoning, not memorization. You will not be asked to recite feed names. You will be asked to judge which action best keeps AI security risk current, prioritized, governed, and communicated. Keep the cycle in mind: define requirements, collect, analyze for relevance, integrate into risk management, act and communicate, measure and improve.
Summary
Continuous monitoring of the AI threat landscape is the mechanism that keeps AI security risk management honest in an environment where attack techniques, dependencies, and regulations change constantly. It is outward-looking, continuous, prioritized by relevance to the organization's AI assets, and valuable only when its outputs flow into the risk register, threat models, controls, and leadership decisions. On the AAISM exam, choose answers that are continuous rather than periodic, integrated rather than isolated, prioritized rather than exhaustive, and governed by accountable management rather than left to tools or vendors.
This guide covers the AAISM (ISACA Advanced in AI Security Management) concept of Continuous Monitoring of the AI Threat Landscape, which sits within the AI Risk Management for Security domain. You will learn why this topic matters, what it actually is, how a mature program operates, and how to recognize and answer exam questions on it.
1. Why Continuous Monitoring of the AI Threat Landscape Is Important
Traditional security risk assessments were often point-in-time exercises: assess annually, document findings, and revisit next year. That model fails for AI for several reasons:
- The attack surface evolves faster than any review cycle. New jailbreak techniques, prompt injection variants, data poisoning methods, model extraction attacks, and adversarial example strategies are published weekly, often in academic preprints before any vendor advisory exists.
- AI systems are probabilistic and opaque. A control that mitigated a threat yesterday may be bypassed tomorrow by a slightly reworded prompt or a new encoding trick. You cannot assume a static control baseline remains effective.
- Dependencies change silently. Foundation models are updated by providers, open-source model weights are replaced, third-party plugins and agents are added, and ML frameworks (PyTorch, TensorFlow, LangChain, etc.) receive security patches. Each change can introduce new vulnerabilities.
- Regulatory and legal exposure is shifting. The EU AI Act, emerging U.S. state laws, sector regulators, and standards bodies (NIST, ISO/IEC 42001) continue to issue new expectations that redefine what counts as an unacceptable risk.
- Adversaries use AI too. Attackers leverage generative AI for phishing, deepfakes, automated vulnerability discovery, and malware generation, meaning the threat environment for all systems, not just AI systems, is changing.
From an ISACA perspective, the key principle is that risk is dynamic. A risk register that is not continuously refreshed with current AI threat intelligence quickly becomes misleading, leading to misallocated controls, false assurance to senior management, and uninformed risk acceptance decisions.
2. What Continuous Monitoring of the AI Threat Landscape Is
Continuous monitoring of the AI threat landscape is the ongoing, structured process of collecting, analyzing, prioritizing, and acting upon information about emerging threats, vulnerabilities, attack techniques, incidents, and regulatory developments relevant to an organization's AI systems, so that risk assessments, threat models, and controls remain current.
It is important to distinguish it from related concepts you will also see on the exam:
- Model performance monitoring (drift detection, accuracy tracking, bias metrics) looks inward at how a deployed model behaves. Threat landscape monitoring looks outward at what adversaries and the environment are doing.
- Security event monitoring (SIEM, SOC alerts) detects attacks currently happening against your infrastructure. Threat landscape monitoring identifies threats that may affect you, often before they are weaponized against you.
- Periodic risk assessment is a scheduled snapshot. Threat landscape monitoring is the continuous feed that keeps the snapshot from going stale between assessments and triggers out-of-cycle reassessments.
Together, these form a layered assurance picture. The exam expects you to understand that threat landscape monitoring is primarily a risk management and governance activity that informs technical teams, not merely a technical tooling exercise.
3. How It Works: Components of an Effective Program
3.1 Define scope and intelligence requirements
Start with the organization's AI inventory and use cases. A company running an internal LLM-based assistant has different intelligence needs than one deploying computer vision in autonomous equipment. Define priority intelligence requirements (PIRs): what threats, to which assets, with what business impact, do we most need to know about? Without this, monitoring produces noise rather than insight.
3.2 Identify and curate sources
Mature programs draw from a blend of sources, for example:
- Threat taxonomies and frameworks: MITRE ATLAS (Adversarial Threat Landscape for AI Systems), OWASP Top 10 for LLM Applications, OWASP Machine Learning Security Top 10, NIST AI RMF and the NIST adversarial ML taxonomy (AI 100-2).
- Vulnerability and advisory feeds: CVE/NVD entries for ML frameworks and libraries, model provider security bulletins, cloud provider AI service advisories, software bill of materials (SBOM) and emerging AI BOM data.
- Incident repositories: the AI Incident Database, AIAAIC repository, and vendor post-mortems that reveal real-world failure modes.
- Research and community: academic preprints, red-team disclosures, security conference talks, bug bounty reports, and practitioner communities.
- Information sharing bodies: sector ISACs/ISAOs, government cyber agencies, and AI-specific sharing initiatives.
- Regulatory and standards trackers: legislative updates, regulator guidance, and standards revisions that change compliance obligations.
- Internal telemetry: findings from your own red teaming, guardrail bypass logs, abuse reports, and model monitoring anomalies, which are often the earliest indicator that an external technique is being used against you.
3.3 Collect and normalize
Automate where possible: subscribe to feeds, use threat intelligence platforms (TIPs) that can ingest AI-specific indicators, and map incoming items to a common taxonomy such as MITRE ATLAS techniques. Normalization allows comparison and trending over time.
3.4 Analyze and triage for relevance
Not every published attack matters to every organization. Analysts assess each item against the AI inventory: Does it affect a model, framework, or provider we use? Does it target a use case we have deployed? Is it theoretical, proof-of-concept, or actively exploited? What is the likely impact and ease of exploitation? This produces a prioritized, contextualized view rather than a raw list.
3.5 Integrate into risk management
This is the step the AAISM exam emphasizes most. Validated intelligence must flow into:
- The AI risk register, updating likelihood, impact, and velocity ratings and adding new risk scenarios.
- Threat models for individual AI systems, so that design-stage assumptions are revisited.
- Control effectiveness assessments, identifying where guardrails, input validation, access controls, or monitoring rules need updating.
- Key risk indicators (KRIs) and risk appetite reviews, so leadership can see whether exposure is trending toward tolerance thresholds.
- Third-party risk management, when the intelligence concerns a model provider, data supplier, or plugin vendor.
3.6 Act and communicate
Define escalation paths and response playbooks: who is notified when a critical AI vulnerability emerges, what interim mitigations are applied, and when an out-of-cycle risk assessment or emergency change is triggered. Communicate in business terms to risk committees and the board, and in technical terms to engineering and SOC teams.
3.7 Measure and improve
Track metrics such as time from public disclosure to internal assessment, percentage of AI assets mapped to current threat taxonomies, number of risk register updates driven by intelligence, and control changes implemented as a result. Review the program itself periodically to ensure sources and PIRs remain aligned with the evolving AI portfolio.
4. Governance Considerations
- Ownership: Typically the CISO or AI risk function owns the process, with contributions from threat intelligence, data science, legal/compliance, and procurement. The exam often tests who should own or approve decisions; the answer is usually management with accountability for risk, informed by specialists.
- Integration with ERM: AI threat intelligence should feed the enterprise risk management framework rather than operate as a silo.
- Documentation: Maintain an audit trail showing how intelligence changed risk ratings and controls. This demonstrates due care to regulators and auditors.
- Three lines model: First line (AI product and engineering teams) applies mitigations, second line (risk and security management) runs and interprets monitoring, third line (internal audit) provides assurance that the process works.
5. How to Answer Exam Questions on This Topic
AAISM questions follow the classic ISACA style: scenario-based, four options, often with two plausible choices where you must pick the best, most important, primary, or first action. Apply these reasoning patterns:
- Favor continuous over periodic. If an option proposes an annual or quarterly review and another proposes ongoing monitoring integrated into risk processes, the continuous option is almost always correct for this concept.
- Favor risk integration over raw collection. Collecting feeds is necessary but not sufficient. The best answer usually involves updating the risk register, threat models, or controls based on the intelligence.
- Favor relevance and prioritization over volume. ISACA values focusing resources on threats that affect the organization's actual AI assets and use cases. An option that says monitor everything equally is a distractor.
- Favor governance and process over tools. If a question asks what is MOST important to establish, the answer is more likely a defined process with ownership and escalation than purchasing a specific platform.
- Favor management accountability. Risk acceptance, appetite changes, and prioritization are management decisions informed by, not delegated to, technical staff or vendors.
- Watch for the FIRST step. When asked what to do first, it is usually to understand the AI inventory and define intelligence requirements, or to assess the relevance of a new threat to the organization, before implementing controls.
- Recognize the outward-looking nature. If a question contrasts drift monitoring or SIEM alerts with threat landscape monitoring, remember that threat landscape monitoring addresses external emerging threats and environmental change.
6. Exam Tips: Answering Questions on Continuous Monitoring of the AI Threat Landscape
Tip 1: Look for the trigger phrase. Scenarios mentioning a newly published jailbreak, a vulnerability in an ML library, a vendor model update, or a new regulation are signaling this concept. The correct response typically involves assessing relevance and updating the risk assessment, not immediately shutting systems down or ignoring the news until the next review.
Tip 2: Eliminate point-in-time answers. Any option implying that a completed risk assessment remains valid until the next scheduled cycle contradicts the dynamic-risk principle and is almost certainly wrong.
Tip 3: Eliminate reactive-only answers. Options that rely solely on responding to incidents after they occur miss the proactive purpose of threat landscape monitoring.
Tip 4: Know the named frameworks. Be able to recognize MITRE ATLAS as the AI-specific adversarial technique knowledge base, OWASP Top 10 for LLM Applications as the common LLM vulnerability list, and NIST AI RMF as the overarching risk framework. Questions may use these as the correct structured source for mapping threats.
Tip 5: Connect to the risk register. When unsure, the option that ties monitoring output to the risk register, KRIs, or control updates is usually the strongest, because it demonstrates the intelligence is being acted upon within governance.
Tip 6: Think about the whole supply chain. The AI threat landscape includes model providers, training data sources, open-source components, plugins, and agents. An option that limits monitoring to in-house models alone is incomplete.
Tip 7: Remember communication. Intelligence that is not communicated to decision-makers and implementers has no value. Options emphasizing timely, audience-appropriate reporting to risk committees and engineering teams reflect mature practice.
Tip 8: Distinguish the roles. Internal audit provides assurance over the process; it does not run the monitoring. Data scientists supply technical interpretation; they do not set risk appetite. The CISO or risk owner coordinates and escalates. Misattributed roles are common distractors.
Tip 9: Apply the business lens. ISACA exams reward answers that translate technical threats into business impact and align responses with risk appetite and organizational objectives.
Tip 10: Practice the reasoning, not memorization. You will not be asked to recite feed names. You will be asked to judge which action best keeps AI security risk current, prioritized, governed, and communicated. Keep the cycle in mind: define requirements, collect, analyze for relevance, integrate into risk management, act and communicate, measure and improve.
Summary
Continuous monitoring of the AI threat landscape is the mechanism that keeps AI security risk management honest in an environment where attack techniques, dependencies, and regulations change constantly. It is outward-looking, continuous, prioritized by relevance to the organization's AI assets, and valuable only when its outputs flow into the risk register, threat models, controls, and leadership decisions. On the AAISM exam, choose answers that are continuous rather than periodic, integrated rather than isolated, prioritized rather than exhaustive, and governed by accountable management rather than left to tools or vendors.
Test mode:
AAISM - Continuous Monitoring of the AI Threat Landscape Example Questions
Test your knowledge of Continuous Monitoring of the AI Threat Landscape
Question 1
An information security manager receives a threat intelligence report describing a newly published indirect prompt injection technique. The technique plants hidden instructions in web pages and shared documents that LLM applications later retrieve. The enterprise runs several LLM-based tools, some built internally and some provided by third parties. What should the security manager do FIRST in response to this report?
Question 2
An information security manager oversees an internal LLM assistant. It uses retrieval-augmented generation over shared document libraries and can send email on users' behalf. The foundation model is hosted by a third-party provider and is reached only through the provider's managed service. In one review cycle, the threat monitoring feed reports four new developments. Which development should MOST prompt a reassessment of this assistant's risk?
Question 3
What is the PRIMARY purpose of continuously monitoring the AI threat landscape within an AI security program?
Unlock Premium Access
Advanced in AI Security Management
- Access to ALL Certifications: Study for any certification on our platform with one subscription
- 3073 Superior-grade Advanced in AI Security Management practice questions
- Unlimited practice tests across all certifications
- Detailed explanations for every question
- AAISM: 5 full exams plus all other certification exams
- 30-Day Money-Back Guarantee: Full refund within 30 days if unsatisfied
- Risk-Free: 7-day free trial with all premium features!
More Continuous Monitoring of the AI Threat Landscape questions
21 questions (total)
Other AI Risk Management concepts
Accountability for Vendor AI Outcomes and IncidentsAdversarial Examples and Evasion AttacksAI Bill of Materials and Component TransparencyAI-Enabled Attacks: Phishing, Deepfakes and MalwareAI Impact Assessments and Regulatory ConformityAI Red Teaming and Adversarial TestingAI Risk Management Life Cycle and Enterprise RiskAI Risk Register and Risk ReportingAI Risk Treatment: Mitigate, Accept, Avoid, TransferAI Vendor Due Diligence and Security AssessmentBias, Fairness and Discrimination RiskData Poisoning and Training Data ManipulationData Usage, Training Rights and Confidentiality ClausesDenial of Service and Resource Abuse Against AIEmbedded AI Features in SaaS and Vendor ProductsExit Strategy and Lock-In for AI ServicesFoundation Model and API Provider DependenciesFourth-Party and Concentration Risk in AI ServicesHallucination and Output Reliability RiskIdentifying AI Risk ScenariosInherent and Residual Risk of AI SystemsInsecure Output Handling and Excessive AgencyLegal, Reputational and Compliance Risk From AILikelihood and Impact Analysis for AI RiskModel Drift and Performance Degradation RiskModel Extraction, Inversion and Membership InferenceMonitoring Triggers for AI Risk ReassessmentOngoing Vendor Monitoring and Right to AuditOpen-Source Models, Datasets and Component RiskPrompt Injection and Jailbreak AttacksRisk Appetite, Tolerance and Thresholds for AIRisk of Agentic AI and Autonomous Decision SystemsRisk of Generative AI and Large Language ModelsRisk Ownership and Acceptance for AI SystemsSecurity Requirements in AI Contracts and SLAsSensitive Information Disclosure From AI SystemsShadow AI Procurement and Unsanctioned ToolsSupply Chain Attacks on Models and ComponentsThe AI Threat Landscape and Threat ActorsThreat Intelligence for AI SystemsThreat Modeling for AI SystemsVendor Model Updates and Change NotificationVerifying Vendor AI Security Claims and AttestationsVulnerability Management and Patching for AI ComponentsVulnerability Scanning and Penetration Testing of AI