Join EY's AI Security practice to safeguard AI/GenAI systems and lead advanced adversarial testing. This role is crucial for protecting AI components throughout their lifecycle and implementing stringent controls within enterprise environments. You'll contribute to building a better working world by ensuring the security of cutting-edge AI technologies.

As a Senior professional in AI & SaaS Product Security, you will focus on Strategic Platforms and AI Security, driving initiatives that build trust through assurance and help clients grow and transform.

Lead comprehensive AI red teaming exercises, including sophisticated prompt injection and data exfiltration simulations. Conduct thorough AI threat modeling across model, data, and application layers to identify vulnerabilities. Design and deploy robust LLM guardrails and RAG security controls to mitigate risks. Perform detailed security reviews of ML pipelines and the entire model lifecycle. Secure APIs and AI inference endpoints within cloud environments, ensuring their integrity. Provide critical support for incident response related to AI-specific threats and exploitation attempts. Integrate security controls seamlessly into AI/ML pipelines and DevSecOps workflows. Collaborate with SOC teams to develop effective detection and monitoring use cases for AI workloads.

Requires 3-6 years of dedicated experience in cybersecurity, application security, or cloud security. Demonstrate hands-on familiarity with AI/GenAI security concepts or practical implementations. Possess experience in vulnerability assessment, red teaming, or threat modeling. Showcase working knowledge of major cloud platforms such as Azure or AWS. Strong understanding of ML/GenAI concepts, including LLMs, embeddings, and pipelines is essential. Knowledge of AI attack surfaces across prompt, model, and data layers is expected. An adversarial mindset with the ability to simulate real-world attack scenarios is key. Familiarity with OWASP Top 10 and LLM vulnerabilities is required. Working knowledge of the MITRE ATT&CK framework is necessary.