Certified AI Security Professional

Prerequisites

• Solid understanding of core cyber security and information security concepts
• Experience or knowledge of threat modelling, identity and access control, incident response, security monitoring, and general security architecture is recommended
• Basic understanding of AI and machine learning concepts, including model training, data pipelines, and API-based AI services

Target audience

• Cyber security and information security professionals
• AI and machine learning engineers
• AI security engineers
• Security architects and consultants
• SOC analysts and incident responders
• Threat intelligence professionals
• Data scientists and AI developers
• AI risk and compliance professionals
• Technology managers and technical leads responsible for AI security
• Professionals responsible for securing AI-enabled applications and services

Certified AI Security Professional Course Content

AI security foundations

• Artificial intelligence concepts and terminology
• Machine learning fundamentals
• Supervised and unsupervised learning
• Deep learning and neural networks
• AI model development and training
• Training, validation, and testing data
• Data pipelines and model lifecycle
• AI inference and deployment
• Understanding the AI technology stack

Security for AI

• How AI security differs from traditional application security
• AI attack surfaces
• Threat actors and adversarial objectives
• AI assets requiring protection
• Confidentiality, integrity, and availability in AI environments
• Model and data integrity
• AI system trust boundaries
• Security across the AI lifecycle
• Threat modelling AI systems

AI architecture and security

• AI infrastructure and platforms
• Model hosting and inference services
• APIs and application integration
• Cloud-based AI services
• Third-party AI dependencies
• AI supply chain considerations
• Security implications of model provenance

Large language model architecture

• LLM architecture and operation
• Transformers and tokenisation
• Model training and fine-tuning
• Foundation models
• Retrieval-Augmented Generation
• Embeddings and vector databases
• LLM application architectures
• System prompts and context
• AI agents and autonomous systems

LLM and generative AI attacks

• Prompt injection
• Direct and indirect prompt injection
• Jailbreaking
• Sensitive information disclosure
• Model manipulation
• Model extraction
• Training data attacks
• Data poisoning
• Adversarial inputs
• Model denial of service
• Insecure output handling
• Excessive agency
• Tool and plugin abuse

AI threat frameworks

• AI attack lifecycle
• MITRE ATLAS
• OWASP LLM Top 10
• Mapping AI attack techniques
• Threat scenarios and attack paths
• Identifying weaknesses across AI architectures

AI red teaming

• Purpose of AI red teaming
• AI security testing methodologies
• Prompt-based security testing
• Testing model boundaries and safeguards
• Identifying exploitable behaviours
• Documenting and communicating findings

Securing AI systems

• Secure AI architecture
• Defence-in-depth for AI
• Input validation
• Output filtering
• Model access controls
• Authentication and authorisation
• Privileged AI operations
• Secrets and credential protection
• Data protection
• Model protection

Securing LLM applications

• Prompt injection mitigations
• Guardrails
• System prompt protection
• Context isolation
• Secure RAG architectures
• Protecting vector databases
• Controlling tool access
• Agent permissions
• Least privilege for AI agents
• Human oversight and approval controls

AI supply chain security

• Third-party models
• Open-source models
• Model repositories
• AI libraries and dependencies
• Training datasets
• Model provenance
• Supplier assurance
• Supply chain compromise
• Dependency and component management

AI security testing

• Vulnerability assessment
• Adversarial testing
• Model security testing
• AI penetration testing
• AI red teaming
• Testing guardrails and controls
• Security assurance before deployment

AI security operations

• AI-specific telemetry
• Logging AI activity
• Monitoring model interactions
• Detecting abnormal AI behaviour
• Security monitoring integration
• SIEM integration
• Detection engineering for AI systems
• Identifying adversarial activity

AI incident response

• AI security incident scenarios
• Detecting AI compromise
• Incident triage
• Investigation
• Containment
• Model isolation
• Recovery
• Evidence collection
• Post-incident analysis

AI-enhanced cyber security

• AI-assisted security monitoring
• AI-assisted threat hunting
• Threat intelligence
• Security analysis
• alware analysis
• AI-assisted reverse engineering
• Opportunities and limitations of defensive AI

AI security management

• AI security governance
• Roles and responsibilities
• AI security policies
• AI risk assessment
• Security assurance
• Regulatory and compliance considerations
• Managing third-party AI risk
• Governance of AI models and agents

Building an AI security programme

• AI asset identification
• AI security strategy
• Secure architecture principles
• AI security controls
• Security testing and red teaming
• Monitoring and incident response
• Supply chain governance
• Continuous assurance
• Integrating AI security into existing cyber security programmes

Exams and assessments

Participants will complete a formal certification exam administered by PECB, post course. Certification fees and the exam voucher are included in the course price. Candidates who do not pass on the first attempt may retake the exam once for free within 12 months of the initial attempt. Knowledge checks, exercises, and quizzes are provided throughout the course to reinforce learning and readiness for the certification exam.

Hands-on learning

• Practical exercises based on real-world AI security scenarios
• Interactive group discussions and case-based simulations
• Workshops on developing and applying AI security techniques