Course Objective & Scope
If an integration granted to an agent for "reading email" can also delete messages, or a database connection opened for reporting carries write access, a single manipulated output can cause serious damage. The course examines the three dimensions of excessive agency (excessive functionality, permissions and autonomy) and the misuse of legitimate tools through tool calling and MCP-based examples.
Participants apply narrowly scoped tool design, schema and parameter validation, action limits, human-in-the-loop approvals and reversibility principles, and finish by producing an agent permission matrix for their own agents.
Who Should Attend?
- Engineers building agents with LangGraph, the OpenAI Agents SDK or similar frameworks
- Teams building automations on platforms such as Copilot Studio and n8n
- Security architects and application security specialists
- Technical leads who approve risk for agent projects
Key Takeaways
- Produce a tool and permission inventory for existing agents
- Redesign tools that carry excessive functionality or permissions
- Write validation and limit rules for tool parameters
- Define approval and rollback flows for risky actions
- Prepare a role- and action-based agent permission matrix
Curriculum
The Three Dimensions of Excessive Agency
- Excessive functionality, permissions and autonomy
- OWASP LLM06 and the tool misuse risk for agentic applications
Tool Misuse Scenarios
- Chaining legitimate tools for unintended purposes
- Cost and resource exhaustion attacks
Secure Tool Design
- Narrow, single-purpose tools and strict schemas
- Parameter validation, rate limits and action quotas
Approval, Rollback and Human Oversight
- Human-in-the-loop points by risk level
- Dry runs, reversible actions and audit logs
Agent Permission Matrix Workshop
- Permission table by agent, tool, action and data
- Risk classification for the organisation's agents
Other Courses in This Category
-
Explains how poisoned documents planted in a knowledge base manipulate retrieval results and LLM answers, and how to prevent this with ingestion pipeline controls, source trust scoring and retrieval anomaly monitoring.
- Live Course
- 5 modules
-
Covers pickle-based code execution, backdoors, licensing and provenance risks in models sourced from hubs such as Hugging Face, and how to manage them with safetensors, ModelScan, model signing and an AI-BOM (CycloneDX).
- Live Course
- 5 modules
-
Teaches how to prevent confused deputy and privilege escalation risks in autonomous agents through workload identity, OAuth 2.0 On-Behalf-Of delegation, dynamic scope attenuation and OPA/Cedar policy enforcement.
- Live Course
- 5 modules
-
Combines STRIDE with MITRE ATLAS and the OWASP Top 10 for LLM and Agentic AI to produce end-to-end AI threat models, trust boundary diagrams and control matrices for LLM, RAG and agent architectures.
- Live Course
- 5 modules
-
Teaches how to plan and execute scenario-driven red team campaigns against LLM, RAG and agent systems using PyRIT, garak and promptfoo, build multi-turn exploit chains and deliver reproducible findings.
- Live Course
- 5 modules
-
Teaches how to design and operate low-latency guardrails across input, retrieval, tool calling and output stages using NeMo Guardrails, Guardrails AI, Llama Guard and deterministic policy engines.
- Live Course
- 5 modules
Get in touch
