Mohit Bansal
Senior Manager, Infrastructure Security
Actions
Mohit Bansal is a Senior Manager of Security Operations Engineering at Webflow, where he leads infrastructure security across AWS, GitHub, and Cloudflare. He has hands-on experience securing cloud-native environments as organizations adopt AI agents and agentic tooling into their production infrastructure.
Prior to Webflow, Mohit spent nearly six years at Okta in progressively senior security engineering roles spanning DevSecOps, Code and Container Security, Supply Chain Security, and Vulnerability Management across SOC 2, FedRAMP, and NIST-regulated environments. He has over a decade of experience building security programs at high-compliance SaaS companies and is an active voice on AI workload security, supply chain risk, and developer-targeting threat campaigns.
Links
Securing the Agentic Pipeline: What Cloud Teams Get Wrong About AI Workload Trust Boundaries
Cloud teams have spent years getting good at securing traditional workloads. IAM roles are scoped, API keys are rotated, pipelines are locked down. Then an AI agent gets deployed, and most of those mental models stop applying.Agents are different. They make decisions autonomously, chain together tools and APIs, and operate with permissions that were designed for humans or simple services, not for systems that can reason, plan, and act across your entire cloud environment. The misconfigurations that result are not careless mistakes. They are the natural outcome of applying old security thinking to a fundamentally new kind of workload.This session surveys the architectural patterns and trust boundary mistakes that appear repeatedly as organizations move AI agents into production on cloud infrastructure. We will cover how agents inherit over-permissioned identities from the services that invoke them, how OAuth scopes and IAM roles designed for point-in-time human actions become dangerous when held by a system that operates continuously, how supply chain risk in AI tooling such as model SDKs and MCP servers introduces new vectors that most cloud security teams are not watching, and how prompt injection turns your own cloud APIs into an attack surface.The goal is not a checklist. The goal is a mental model shift: understanding why agents require a different security contract than the workloads you already know how to protect, and what the foundational principles of that contract look like in practice.
Securing the Agentic Pipeline: What Cloud Teams Get Wrong About AI Workload Trust Boundaries
Cloud teams have spent years getting good at securing traditional workloads. IAM roles are scoped, API keys are rotated, pipelines are locked down. Then an AI agent gets deployed, and most of those mental models stop applying.Agents are different. They make decisions autonomously, chain together tools and APIs, and operate with permissions that were designed for humans or simple services, not for systems that can reason, plan, and act across your entire cloud environment. The misconfigurations that result are not careless mistakes. They are the natural outcome of applying old security thinking to a fundamentally new kind of workload.This session surveys the architectural patterns and trust boundary mistakes that appear repeatedly as organizations move AI agents into production on cloud infrastructure. We will cover how agents inherit over-permissioned identities from the services that invoke them, how OAuth scopes and IAM roles designed for point-in-time human actions become dangerous when held by a system that operates continuously, how supply chain risk in AI tooling such as model SDKs and MCP servers introduces new vectors that most cloud security teams are not watching, and how prompt injection turns your own cloud APIs into an attack surface.The goal is not a checklist. The goal is a mental model shift: understanding why agents require a different security contract than the workloads you already know how to protect, and what the foundational principles of that contract look like in practice.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top