AI Security: Securing LLM and Agentic AI Applications

Large language models (LLMs) and agentic AI applications are increasingly being incorporated into enterprise software, business workflows and connected systems.

While these technologies create new opportunities for automation and intelligent applications, they also introduce security challenges that differ from those found in traditional software systems.

AI applications may process untrusted inputs, access sensitive information, interact with external tools and perform actions across connected environments. These capabilities require careful attention to security architecture, access controls and operational oversight.

This NobleProg hybrid event explores security considerations for LLM-powered applications and agentic AI systems, focusing on common threats, defensive approaches and responsible deployment practices.

Participants can attend onsite in Andheri East, Mumbai, or join live online, with both attendance options running simultaneously.

Why Attend?

Securing AI applications requires understanding how models, prompts, retrieved information, external tools and application permissions interact.

This event explores how technical teams can:

• Recognise security risks associated with LLM-powered applications.
• Understand prompt injection and untrusted input threats.
• Identify potential sensitive data exposure risks.
• Evaluate security concerns in retrieval-augmented generation (RAG) workflows.
• Understand tool-use risks in agentic AI applications.
• Explore authentication, authorisation and least-privilege principles.
• Consider human approval mechanisms for sensitive agent actions.
• Apply security thinking to AI application design and deployment.

Key Topics

1. Introduction to LLM and Agentic AI Security

Explore the security characteristics of LLM-powered and agent-based systems.

Consider how AI application architectures introduce new trust boundaries involving users, models, prompts, data sources, tools and external services.

2. Prompt Injection and Untrusted Inputs

Understand how malicious or misleading instructions embedded in user inputs, retrieved documents or external content can influence AI application behaviour.

Explore defensive concepts such as:

• Separating trusted instructions from untrusted content.
• Validating inputs and tool arguments.
• Limiting application permissions.
• Reducing the impact of manipulated model outputs.
• Testing applications against prompt injection scenarios.

3. Sensitive Data Protection

Examine risks associated with confidential information in AI workflows, including:

• Exposure of sensitive information through prompts or outputs.
• Inappropriate access to internal documents.
• Data handling and retention considerations.
• Protecting credentials and secrets.
• Applying appropriate access restrictions.

4. Security in RAG Applications

Explore security considerations for retrieval-augmented generation systems.

Potential areas include document trustworthiness, retrieval permissions, information leakage and the handling of malicious content within knowledge sources.

5. Agentic AI and Tool Security

Understand how AI agents introduce additional risks when interacting with tools, APIs and connected enterprise systems.

Explore considerations such as:

• Tool permission boundaries.
• Authentication and authorisation.
• Untrusted tool responses.
• Unintended or unauthorised actions.
• Tool invocation validation.
• Least-privilege access.

6. Human Approval and Agent Governance

Explore the role of human-in-the-loop approval for sensitive or high-impact actions.

Consider approval checkpoints, escalation procedures, audit trails and accountability when AI agents interact with business systems.

7. Threat Modelling and Security Testing

Examine how teams can approach security assessments for AI applications.

Potential topics include:

• Identifying AI application assets and trust boundaries.
• Mapping potential attack surfaces.
• Evaluating misuse scenarios.
• Testing security controls.
• Reviewing AI-specific application risks.
• Monitoring and investigating unexpected behaviour.

8. Secure AI Application Deployment

Explore security principles relevant to deploying LLM and agentic AI applications in organisational environments.

Consider configuration management, access restrictions, logging, monitoring, incident response and ongoing security review.

These are proposed topic areas based on the event title. Final curriculum coverage, security frameworks, demonstrations and practical exercises require confirmation against the approved course agenda.

Who Should Attend?

This event is relevant to:

• Cybersecurity professionals
• Application security engineers
• AI and machine learning engineers
• LLM application developers
• Software engineers building AI-powered systems
• Security architects and solution architects
• DevSecOps professionals
• Enterprise AI platform teams
• Technical professionals responsible for AI security and governance

Delivery Format Hybrid — Onsite and Live Online

Participants can attend physically at the Mumbai venue or join the same event remotely through live online participation.

About NobleProg

NobleProg provides professional training in technical and business-related subjects, supporting individuals and organisations in developing relevant knowledge and skills.

Registration

Register your interest in AI Security Training for LLM and Agentic AI Applications in Mumbai.

Choose onsite participation in Mumbai or live online attendance when the event schedule and registration options are confirmed. Contact us at training@nobleprog.in