Webinar Description
Key Takeaways
- Addresses security gaps in AI agents that perform critical business actions such as financial transactions, code modifications, and data access.
- Examines why protecting the full agentic path (LLMs, MCP servers, tools, and downstream APIs) requires a different approach than traditional model-level security.
- Relevant for CISOs, security architects, and technology leaders in regulated industries including finance and healthcare.
- Hosted by Salt Security, focusing on agentless visibility and runtime protection for enterprise AI deployments.
About the Event
This virtual webinar from Salt Security examines the security challenges that emerge when enterprises deploy AI agents to perform consequential business operations. The session distinguishes between securing what AI models say (prompt injection and model-level threats) and securing what AI agents actually do when they interact with production systems, APIs, and sensitive data.
The Agentic Attack Surface
AI agents powered by large language models increasingly execute real business actions: moving funds, accessing sensitive records, modifying code repositories, and sending communications on behalf of users. Each of these actions involves a chain of components, including the LLM itself, MCP servers that coordinate agent behaviour, various tools, and downstream APIs that connect to enterprise systems.
The webinar argues that most current security programmes concentrate on model-level concerns while leaving the broader agentic path unprotected. This creates a blind spot where agents can take harmful actions even when the underlying model appears secure. The session explores why single-layer defences are insufficient and how attackers might exploit gaps between components in the agentic chain.
Salt Security’s Approach
The webinar introduces Salt Security’s methodology for addressing these risks through agentless visibility and governance across the full agentic path. The approach aims to provide runtime protection without requiring organisations to modify their existing architecture. The session covers how to achieve unified security across LLMs, MCP servers, tools, and downstream API integrations.
Who Should Attend
The webinar is designed for security professionals responsible for protecting AI deployments, including CISOs, security architects, and security engineers. Product managers and developers building AI-powered applications will find relevant technical context. The content is particularly applicable to organisations in regulated industries such as finance and healthcare, where AI agents may interact with sensitive data and compliance-critical systems. Senior decision-makers responsible for enterprise risk posture and compliance strategy are also part of the intended audience.