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 21 hours
Course Outline
Foundations: Threat Models for Agentic AI
- Identifying agentic threats, including misuse, escalation, data leakage, and supply-chain risks.
- Profiling adversaries and understanding attacker capabilities specific to autonomous agents.
- Mapping assets, trust boundaries, and critical control points associated with agents.
Governance, Policy, and Risk Management
- Implementing governance frameworks for agentic systems, covering roles, responsibilities, and approval gates.
- Designing policies for acceptable use, escalation rules, data handling, and auditability.
- Addressing compliance considerations and collecting evidence for audits.
Non-Human Identity and Authentication for Agents
- Designing agent identities using service accounts, JWTs, and short-lived credentials.
- Applying least-privilege access patterns and just-in-time credentialing.
- Managing the identity lifecycle, including rotation, delegation, and revocation strategies.
Access Controls, Secrets, and Data Protection
- Applying fine-grained access control models and capability-based patterns for agents.
- Managing secrets, ensuring encryption in transit and at rest, and practicing data minimization.
- Safeguarding sensitive knowledge sources and PII from unauthorized agent access.
Observability, Auditing, and Incident Response
- Designing telemetry for agent behaviour, including intent tracing, command logs, and provenance.
- Integrating with SIEMs, setting alerting thresholds, and maintaining forensic readiness.
- Developing runbooks and playbooks for agent-related incidents and containment.
Red-Teaming Agentic Systems
- Planning red-team exercises, defining scope, rules of engagement, and safe failover procedures.
- Utilising adversarial techniques such as prompt injection, tool misuse, chain-of-thought manipulation, and API abuse.
- Conducting controlled attacks to measure exposure and impact.
Hardening and Mitigations
- Implementing engineering controls like response throttles, capability gating, and sandboxing.
- Applying policy and orchestration controls, including approval flows, human-in-the-loop mechanisms, and governance hooks.
- Employing model and prompt-level defenses such as input validation, canonicalization, and output filters.
Operationalizing Safe Agent Deployments
- Adopting deployment patterns such as staging, canary, and progressive rollout for agents.
- Establishing change control, testing pipelines, and pre-deploy safety checks.
- Fostering cross-functional governance through playbooks for security, legal, product, and operations teams.
Capstone: Red-Team and Blue-Team Exercise
- Executing a simulated red-team attack against a sandboxed agent environment.
- Defending, detecting, and remediating as the blue team using established controls and telemetry.
- Presenting findings, remediation plans, and policy updates.
Summary and Next Steps
Requirements
- A robust foundation in security engineering, system administration, or cloud operations.
- Working familiarity with AI/ML concepts and the behaviour of large language models (LLMs).
- Practical experience with identity and access management (IAM) and secure system design.
Audience
- Security engineers and red-teamers.
- AI operations and platform engineers.
- Compliance officers and risk managers.
- Engineering leads accountable for agent deployments.
Testimonials (1)
inventory and identifying the different risk exposures within AI
