The Cloud & Smart Industries Group (CSIG) at Tencent is seeking an Agent Security Expert to define the technical architecture for next-generation agent security products designed to protect enterprise AI workloads. This role sits at the intersection of application security, AI safety, and systems engineering. Responsibilities include: Defining end-to-end technical architecture for real-time agent defense; Designing runtime guardrails for prompt injection, unauthorized tool-call interception, behavioral drift, and credential leak prevention; Architecting security capabilities against tool poisoning and privilege escalation; Building automated agent red-teaming pipelines; Partnering with cross-functional teams to translate attack patterns into security features; Mentoring junior engineers. Required Qualifications: 8+ years of hands-on experience in security or systems engineering with a track record of shipping production components; Deep expertise in at least two areas: application security, cloud-native security, IAM, or offensive/defensive security operations; Proficiency in at least two of Go, Python, Rust, or C++ with experience in high-concurrency services; Strong knowledge of the LLM/Agent stack including RAG, function calling, and MCP protocol. Preferred Qualifications: Experience researching or shipping enterprise security products (EDR, CNAPP, AI-SPM); Familiarity with sandboxing, eBPF, or container/virtualization isolation.
For full details about this role, click “Apply on company site” to view the employer’s original posting.
A focused preparation guide, built around this job’s responsibilities and requirements.
✦ AI-generated guide
Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.
01
Understand the role
What matters most
Architecting Real-Time Agent Defense Systems
The primary objective is to design the end-to-end technical architecture for securing enterprise AI workloads. This requires balancing the trade-offs between detection coverage, system latency, and operational overhead, ensuring that security guardrails do not impede the performance of autonomous agents.
Implementing Runtime Guardrails and Tool Security
You must develop robust mechanisms to intercept unauthorized tool calls, prevent prompt injection, and mitigate supply-chain risks like tool poisoning. This involves securing the MCP protocol and managing cross-origin privilege escalation to ensure agents operate within defined security boundaries.
Building Automated Red-Teaming Pipelines
A core responsibility is creating automated testing infrastructure to continuously discover vulnerabilities and generate adversarial test cases. This pipeline is essential for hardening defenses against evolving attack patterns before they reach production environments.
Likely challenges to think through
Balancing Security Depth with High-Concurrency Requirements
The role requires building security services that operate at scale. The challenge lies in implementing complex inspection logic—such as behavioral drift detection—without introducing significant latency into high-concurrency AI agent workflows. Prepare by discussing your experience optimizing performance in security-sensitive, high-throughput systems.
Securing Evolving AI Attack Surfaces
The threat landscape for LLM-based agents, including transitive trust abuse and prompt injection, is highly dynamic. The difficulty is designing a product that remains effective against novel attack vectors. Prepare by demonstrating a methodology for translating emerging threat intelligence into modular, adaptable security features.
03
Practice your answers
Technical practice
Architecting Real-Time Agent Defense
What to coverExplain how you would design a low-latency security layer that intercepts agent tool calls without significantly impacting application performance. Discuss the trade-offs between synchronous blocking checks and asynchronous monitoring in high-concurrency environments.
Mitigating Prompt Injection and Behavioral Drift
What to coverDescribe your approach to implementing runtime guardrails that detect prompt injection and behavioral drift in autonomous agents. Focus on how you would distinguish between legitimate agent reasoning and malicious manipulation of the model's instructions.
Securing the MCP and Tool-Chain Supply Chain
What to coverHow would you architect security controls to prevent tool poisoning and transitive trust abuse within an MCP-based agent framework? Address how you would validate the integrity of tools and manage cross-origin privilege escalation.
Building Automated Red-Teaming Pipelines
What to coverOutline the components required for an automated red-teaming pipeline for AI agents. Explain how you would generate adversarial test cases and integrate these findings into a continuous hardening process for production models.
High-Concurrency Security Services with Go or Rust
What to coverGiven your experience with Go or Rust, how do you handle memory safety and concurrency when building security detection pipelines? Provide an example of how you have optimized a high-throughput service to minimize latency.
Isolation Techniques for AI Workloads
What to coverCompare the use of eBPF, container isolation, and sandboxing for securing enterprise AI workloads. Explain when you would choose one over the other based on the need for performance versus strict security boundaries.
Credential Leak Prevention in LLM Interactions
What to coverHow would you design a real-time system to detect and prevent the leakage of sensitive credentials or PII during LLM function calls? Discuss the challenges of inspecting dynamic, unstructured data in real-time.
Integrating AI-SPM into Enterprise Security
What to coverHow does AI Security Posture Management (AI-SPM) differ from traditional CNAPP or EDR solutions? Explain how you would adapt existing security frameworks to address the unique attack surfaces of LLM-integrated applications.
Behavioural practice
Mentoring Junior Engineers in Security-First Design
What to coverDescribe a time you mentored a junior engineer on security best practices. How did you balance the need for rapid development with the requirement for robust, secure architecture?
Cross-Functional Collaboration on Security Features
What to coverTell me about a time you had to translate complex attack patterns into actionable security features for product and engineering teams. How did you ensure the technical requirements were understood and prioritized?
Navigating Technical Trade-offs
What to coverProvide an example of a project where you had to make a difficult trade-off between detection coverage and system performance. How did you evaluate the risks and justify your final architectural decision?
Handling Security Incidents in Production
What to coverDescribe a situation where you identified a vulnerability in a production system. What was your process for containment, root cause analysis, and ensuring the vulnerability was permanently mitigated?
02
Sharpen your knowledge
Topics to prepare
AI Agent Attack Surfaces
Review the specific security risks associated with autonomous agents, including prompt injection, tool poisoning, and transitive trust abuse. Exercise: Map out a threat model for an agent using function calling and identify where unauthorized tool-call interception could be implemented. This is critical for designing the runtime guardrails mentioned in the job description.
Supply Chain Security for AI
Study security architectures for the Model Context Protocol (MCP) and tool-chain integrations. Exercise: Research how to detect and prevent privilege escalation when an agent is granted access to external APIs or internal tools. This is directly relevant to the requirement for architecting MCP and tool-chain security.
Skills to revise
High-Concurrency Systems Programming
Practice building high-performance services in Go or Rust. Exercise: Design a prototype for a real-time security proxy that intercepts and inspects agent traffic with minimal latency impact. This addresses the need for building high-concurrency security services.
Automated Red-Teaming Pipelines
Develop a framework for automated adversarial testing of LLM applications. Exercise: Create a script that generates adversarial test cases to probe for prompt injection vulnerabilities in a sample agent workflow. This aligns with the requirement to build automated agent red-teaming pipelines.
Runtime Isolation Techniques
Gain proficiency in sandboxing and container isolation technologies like eBPF. Exercise: Configure a containerized environment to restrict an agent's network and file system access using eBPF-based policies. This is a preferred qualification for securing enterprise AI workloads.
04
Your preparation checklist
Map AI Agent Attack Surfaces to Security Controls
Prepare a technical summary or whitepaper outlining how you would architect runtime guardrails for LLM-based agents. Focus on mitigating prompt injection, tool poisoning, and unauthorized function calls, specifically referencing how you would handle the MCP protocol and RAG-based data retrieval.
Design a High-Concurrency Security Pipeline
Draft a system architecture diagram for a real-time agent defense service. Demonstrate how you would balance low-latency requirements with deep packet or prompt inspection, using your preferred language (Go, Rust, Python, or C++) and incorporating eBPF or container isolation techniques.
Develop an Automated Red-Teaming Strategy
Outline a methodology for building an automated red-teaming pipeline that generates adversarial test cases for AI agents. Explain how you would integrate this into a CI/CD workflow to ensure continuous hardening against privilege escalation and behavioral drift.
Prepare Case Studies on Production Security Systems
Select two past projects where you shipped production-grade security components (e.g., EDR, CNAPP, or IAM systems). Be ready to discuss the trade-offs you made between detection coverage, system performance, and operational overhead.
Formulate Strategic Questions for the Team
Prepare questions regarding the current maturity of their agent security product, the specific agent frameworks they prioritize, and how the team balances rapid AI innovation with the stringent security requirements of enterprise cloud environments.
Availability last checked: 18 Oct 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy
Share this opportunityKnow someone who would be a great fit?