Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Duration 14 hours
Course Outline
The AI Threat Landscape
- Distinctions in AI security: addressing non-deterministic behavior, opaque reasoning processes, and prompts as critical attack surfaces
- Classifying threats: differentiating between training-time, inference-time, and supply chain attack vectors
- The adversary model: identifying entities targeting AI systems and their strategic motivations
OWASP Top 10 for LLM Applications
- Prompt injection: analyzing direct and indirect attack vectors
- Risks associated with insecure output handling and cross-plugin request forgery
- Training data poisoning and vulnerabilities within the model supply chain
- Denial of service risks, sensitive information disclosure, and excessive agency concerns
- Practical laboratory exercise: exploiting each OWASP category against a target application
Prompt Injection and Jailbreak Red Teaming
- Categorizing injection techniques: direct, indirect, multi-turn, and multi-modal approaches
- Automating red-teaming efforts using Giskard, Garak, and custom fuzzing tools
- Classifying jailbreak attempts and evaluating defensive measures
- Constructing a red-team harness for continuous LLM security assessment
Model-Level Attacks and Defenses
- Model extraction: methods for retrieving model weights and functionality via API queries
- Membership inference: techniques for determining if specific data was included in the training set
- Adversarial examples: analyzing perturbations designed to deceive classifiers and embeddings
- Data poisoning: strategies for corrupting training data to induce backdoors or degrade performance
Input and Output Security Controls
- Input sanitization methods that extend beyond traditional web security defenses
- Output filtering: mitigating toxicity, PII leakage, and hallucinated code execution
- Implementing guardrails as core security infrastructure: utilizing NeMo, Guardrails AI, and custom policies
- Enforcing structured outputs as a security boundary for government systems
AI Supply Chain Security
- Model provenance: verifying the authenticity and integrity of model artifacts
- Dependency scanning for ML frameworks and proprietary model formats
- Secure model serving: applying sandboxing, network isolation, and least-privilege access controls
- Vetting fine-tuned and community-sourced models for embedded malware
Operational Security for AI Systems
- Implementing access control for model endpoints, vector stores, and agent tools
- Maintaining audit logs for every model interaction and decision-making process
- Incident response for AI-specific breaches: addressing scenarios where the model itself is compromised
- Integrating continuous security testing into CI/CD pipelines for ML operations
Building an AI Security Program
- Establishing an AI security maturity model and strategic roadmap
- Integrating AI security into existing AppSec and cloud security programs
- Aligning with governance frameworks and emerging regulations for AI systems
- Creating and maintaining an organizational AI security playbook
Requirements
- Experience deploying ML models or LLM applications in production environments.
- Familiarity with core security concepts, including authentication, authorization, and threat modeling.
- Proficiency in Python for executing adversarial testing exercises.
Audience
- Security engineers expanding their scope to include AI/ML threat surfaces.
- ML engineers responsible for ensuring model safety and robustness.
- Red team members integrating AI systems into their testing scope for government.
