AI Security courses

• Understand the security landscape of machine learning, generative AI, large language models, Retrieval-Augmented Generation systems, and AI agents, including how their security requirements differ from conventional software systems.
• Explore how AI introduces new attack surfaces across data, models, prompts, applications, infrastructure, integrations, external tools, and users.
• Identify common AI threats, vulnerabilities, misuse patterns, and security failures that can compromise confidentiality, integrity, availability, reliability, or organizational trust.
• Build a foundational framework for securing AI systems throughout their lifecycle, from design and development through deployment, operation, monitoring, and retirement.

• Learn how to systematically identify and analyze threats across AI architectures, workflows, integrations, and operational environments before they become exploitable weaknesses.
• Map critical assets, trust boundaries, data flows, models, APIs, tools, identities, infrastructure components, and external dependencies within an AI system.
• Analyze realistic attack scenarios involving models, users, applications, retrieved content, external services, and third-party components.
• Build practical AI threat models and prioritize preventive, detective, and corrective security controls according to likelihood, impact, exposure, and business risk.

• Understand direct and indirect prompt injection, jailbreaks, instruction manipulation, adversarial inputs, and other techniques used to influence or bypass intended LLM behavior.
• Analyze how attackers can exploit system prompts, external content, retrieved documents, user inputs, conversation context, and application instructions to manipulate model behavior.
• Test LLM applications against representative attack patterns, control bypasses, unexpected model behaviors, and malicious contextual inputs.
• Implement layered defenses using isolation, validation, permissions, least privilege, content handling controls, monitoring, and secure application architecture rather than relying on prompts alone.

• Identify security risks across document ingestion, transformation, embeddings, vector databases, indexing, retrieval, context construction, and generation in RAG systems.
• Understand knowledge-base poisoning, malicious documents, unauthorized retrieval, retrieval manipulation, sensitive-information leakage, and compromised source provenance.
• Implement document-level permissions, identity-aware access, metadata filtering, authorization checks, secure ingestion, and controlled retrieval strategies.
• Build RAG architectures that preserve confidentiality, integrity, provenance, access boundaries, and trust throughout the complete knowledge-to-generation pipeline.

• Understand the security implications of giving AI agents access to APIs, databases, enterprise applications, files, code execution, communication systems, and other external tools.
• Address excessive agency, insecure tool execution, permission escalation, unintended actions, unsafe chaining of capabilities, and exploitation of agent decision-making.
• Implement least privilege, authentication, authorization, sandboxing, execution constraints, validation, transaction controls, and human approval mechanisms.
• Design agent architectures that remain controlled, observable, and recoverable even when model behavior is probabilistic, unexpected, or influenced by malicious inputs.

• Protect training, fine-tuning, evaluation, retrieval, and inference data against poisoning, manipulation, leakage, corruption, and unauthorized access.
• Understand model theft, tampering, malicious or compromised models, unsafe dependencies, vulnerable libraries, poisoned datasets, and untrusted artifacts.
• Assess security risks associated with open-source models, third-party APIs, external datasets, software libraries, model repositories, cloud services, and AI providers.
• Establish security controls across the complete AI development and supply-chain lifecycle, including provenance, integrity verification, access management, dependency governance, and vendor assurance.

• Design systematic security assessments that simulate realistic adversarial behavior against AI models, applications, retrieval systems, agents, and supporting infrastructure.
• Evaluate systems for prompt injection, jailbreaks, sensitive-data exposure, harmful behavior, tool misuse, authorization bypasses, control failures, and other application-specific attack scenarios.
• Build repeatable adversarial test suites and structured red-team scenarios that cover normal operation, edge cases, malicious inputs, and chained attacks.
• Transform discovered vulnerabilities into prioritized remediation actions, regression tests, measurable security requirements, and continuous adversarial testing processes.

• Integrate security into AI engineering from architecture and implementation through model integration, testing, deployment, operation, and continuous improvement.
• Establish secure coding, model-access controls, secrets management, dependency management, artifact integrity, environment isolation, and secure deployment practices.
• Introduce AI-specific security checks, adversarial tests, policy validation, dependency scanning, and control verification into CI/CD, MLOps, and LLMOps pipelines.
• Build secure-by-design development processes that continuously identify, prioritize, remediate, and prevent AI vulnerabilities throughout rapid development cycles.

• Monitor prompts, outputs, retrieval activity, tool calls, model behavior, identities, permissions, application events, and infrastructure signals for suspicious or abnormal patterns.
• Define security indicators, detection logic, thresholds, and alerts for AI-specific attacks, misuse, data exposure, anomalies, policy violations, and control failures.
• Build incident-response procedures for compromised AI applications, malicious interactions, unsafe autonomous actions, information leakage, and unexpected model behavior.
• Establish logging, traceability, investigation, containment, recovery, evidence preservation, root-cause analysis, and post-incident improvement processes for AI security events.

• Design an end-to-end enterprise security architecture covering AI models, applications, RAG systems, agents, data, infrastructure, APIs, identities, integrations, vendors, and users.
• Establish security policies, ownership, control requirements, risk assessments, architecture reviews, approval processes, exception handling, and assurance mechanisms for enterprise AI.
• Align AI security with cybersecurity, privacy, data governance, enterprise risk management, software security, incident response, and broader AI governance programs.
• Build a scalable AI security roadmap that enables organizations to expand AI adoption while maintaining resilience, appropriate control, regulatory readiness, and stakeholder trust.

See also: AI Governance courses →