Senior Privacy Security Engineer (AI Red Teaming) - PDPO (Singapore)

The Privacy and Data Protection Office (PDPO) leads, supervises, and empowers TikTok’s global privacy efforts, ensuring that privacy is protected across our products, platforms, and technologies in an accountable and industry-leading manner. As experts in privacy risk assessment and mitigation, we work closely with engineering, product, and cross-functional teams to integrate effective privacy safeguards and technical controls throughout the product and system lifecycle.

As Large Language Models (LLMs) and AI Agents continue to evolve, we explore emerging privacy and security challenges through cutting-edge research, adversarial testing, and automation. Our mission is to identify and mitigate privacy risks in AI systems and continuously advance TikTok’s privacy protection capabilities.

Responsibilities

• Model Privacy Security Assessment: Conduct in-depth privacy security assessments of LLMs deployed across international short-video products and other applications. Evaluate privacy risks across model training data, model outputs, inference services, and user data interaction pipelines, including model memorization, training data leakage, and membership inference. Build and continuously improve systematic and scalable model privacy evaluation frameworks.
• LLM Agent Privacy Security Red Teaming: Develop threat models for LLM Agent systems and conduct adversarial testing across attack surfaces such as prompt injection, unauthorized tool or plugin invocation, context and memory leakage, and cross-session data contamination. Proactively identify privacy and data protection vulnerabilities and work with business and engineering teams to strengthen Agent privacy safeguards.
• Automated Red Team Agent Development: Design and develop automated red teaming agents and evaluation frameworks to identify privacy vulnerabilities in Models and Agents at scale. Build sustainable and repeatable testing capabilities to improve assessment automation, testing efficiency, and risk coverage.
• Frontier AI Privacy Security Research: Conduct research into emerging AI privacy and security challenges, including model memorization and unlearning, training data extraction, membership inference attacks, and privacy-preserving fine-tuning. Explore novel attack techniques and risk assessment methodologies to advance the team’s technical capabilities and expertise in LLM privacy security.

Qualifications

Minimum Qualifications:
- Understanding of Large Language Models (LLMs) and Agents, with familiarity with the fundamental processes of LLM training and inference, Prompt Engineering, RAG, Function Calling / Tool Use, as well as the underlying principles of common Agent orchestration frameworks such as LangChain and AutoGen.
- Familiar with common LLM / Agent security risks and offensive and defensive techniques, including Prompt Injection, Jailbreaks, training data extraction, and membership inference attacks. Proficiency in relevant evaluation tools, frameworks, and red teaming methodologies is preferred.
- Have hands-on experience identifying privacy and security issues in Models or Agents, with the ability to independently carry out the end-to-end process from threat modeling and attack construction to privacy impact assessment. Demonstrate a systematic and comprehensive understanding of AI system attack surfaces.
- Familiar with the principles of common web application frameworks, cloud service architectures, and database storage systems. Practical experience in penetration testing in production environments or building security defense systems is preferred.
- Demonstrate strong logical thinking, teamwork, communication, and execution skills

Preferred Qualifications

• Experience participating in HW or CTF competitions, with strong performance or notable results.
• Publications in the field of privacy and security or experience presenting at relevant academic or industry conferences.
• Hands-on experience discovering and submitting high- or critical-severity privacy security vulnerabilities through Security Response Centers (SRCs).