[EJV] Lead AI Safety & Security Engineer

We are looking for a Lead AI Safety & Security Engineer to own safety evaluation and security testing across Bosch's growing portfolio of AI-enabled products and internal tooling — large language model deployments and agentic AI systems (including Model Context Protocol – style tool-calling integrations), and the enterprise services they connect to. You will lead both sides of the same coin: whether an AI system behaves safely (does it refuse harmful requests, avoid unsafe recommendations, stay within its intended scope) and whether it is secure (can it be manipulated via prompt injection, tricked into misusing its tools, or made to leak data it shouldn't). This is a cross-business-unit mandate, not a single product line — you will build the practice largely from the ground up.

• Design and run safety evaluations for LLM-based features and AI agents: red-teaming for harmful, biased, or unsafe outputs; jailbreak and guardrail-bypass testing; systematic benchmarking against known safety failure modes.
• Design and run security testing for the same systems: prompt injection (direct and indirect), excessive agency, tool/plugin abuse, and confused-deputy risks in agentic and MCP-style tool-calling architectures.
• Build and maintain a threat model covering the full path from user input through model reasoning to tool execution — identifying where a safety failure and a security failure are really the same underlying gap.
• Verify human-in-the-loop and guardrail controls are architecturally sound wherever an AI agent can take a consequential action (a write to a business system, an automated communication, a code change) — not just present in a UI that can be talked around.
• Scope, brief, and review the output of external red-teaming/pentest vendors; write test specifications and acceptance criteria for AI-specific engagements.
• Track and apply emerging frameworks — OWASP Top 10 for LLM Applications, OWASP Top 10 for Agentic Applications, the NIST AI Risk Management Framework, ISO/IEC 42001, and the EU AI Act — translating new research into concrete internal test requirements before it becomes an incident.
• Partner with AI/ML engineering teams across business units to get safety and security reviewed into system design, not only tested after the fact.
• Produce evidence and documentation supporting AI governance and compliance needs (ISO/IEC 42001 AI management system, EU AI Act risk assessments, internal responsible-AI review processes).
• Build test tooling as needed: adversarial-prompt harnesses, automated jailbreak/red-team pipelines, and MCP/tool-schema fuzzers.
• Mentor engineers on safe and secure AI system design, and grow the practice as it scales across more teams and products.