Course Objective & Scope
When an agent is granted permission to read emails, query databases and call external APIs, a hidden instruction inside a retrieved document (indirect prompt injection) can trigger data exfiltration or unauthorised actions. This programme covers identity, authorisation and runtime security for agent systems in depth.
Participants analyse excessive agency, tool misuse and data exfiltration scenarios, implement delegated user authorisation with OAuth On-Behalf-Of (OBO), policy engines (OPA/Cedar), code/browser sandboxes and runtime guardrails, and produce an enterprise agent security checklist.
Who Should Attend?
- Application security (AppSec), cybersecurity and IAM engineers
- AI and backend architects building agent platforms
- Technology risk and compliance specialists
- Prerequisite: basic web/API security and familiarity with LLM/agent architectures
Key Takeaways
- Map attack surfaces by performing threat modelling on agent architectures
- Issue dynamic, least-privilege permissions using delegated user identity (OBO) instead of shared service accounts
- Block unauthorised tool calls triggered by indirect prompt injection at the policy layer
- Isolate code-executing and web-browsing agents at the network and container level
- Implement immutable audit logs for forensics and runtime kill switches
Curriculum
Agentic Threat Modelling and OWASP Risks
- Excessive agency, indirect prompt injection and the confused deputy problem
- Attack chains triggered by untrusted content from retrieved data sources
Identity and Delegated Authorisation (IAM & OAuth OBO)
- Agent identity, acting On-Behalf-Of the user and task-scoped short-lived tokens
- Deterministic tool- and argument-level policy enforcement with OPA and Cedar
Runtime Guardrails and Exfiltration Prevention
- Input/output filtering, PII masking and outbound domain/URL allowlists
- Separating untrusted data ingestion from external communication tools
Sandbox Isolation and Safe Execution
- MicroVM and container isolation for code-running and file-handling agents
- Network egress rules, resource quotas and secure secret injection
Audit Trails, Red Teaming and Incident Response
- Immutable audit logs: which user, agent and context invoked which tool
- Agent red teaming scenarios, anomaly detection and kill switch mechanisms
Other Courses in This Category
-
Foundational programme on how LLM-based agents work: model, tools, memory and the control loop; how agents differ from chatbots and workflow automation, and how to pick the right first use case for your organisation.
- Live Course
- 6 modules
-
Agentic AI for executives: impact on business models, processes and organisation, function-by-function use cases, investment and ROI assessment, operational risk, governance and redesigning teams that work alongside agents.
- Live Course
- 5 modules
-
Hands-on programme on how LLMs use APIs, databases and internal systems via function calling / tool calling: tool schema design, argument validation, error handling and a working agent built on OpenAI, Claude and Gemini APIs.
- Live Course
- 6 modules
-
Method for turning business processes into agentic workflows: separating deterministic steps from LLM decisions, prompt chaining, routing, parallelisation and evaluator–optimizer patterns, implemented in n8n or Power Automate.
- Live Course
- 6 modules
-
Developer programme on building autonomous agents with LangGraph, the OpenAI Agents SDK and the Claude Agent SDK: state management, tool integration, loop control, tracing and a deployment-ready agent for an organisational case.
- Live Course
- 6 modules
-
Technical programme on how agents plan and execute multi-step tasks: ReAct, plan-and-execute, task decomposition, reflection and replanning, using reasoning models, and loop termination strategies.
- Live Course
- 6 modules
Get in touch
