Senior AI Security Engineer — Agentic Red Team

Google LLC; Zürich; Hybrid; CHF 144.000 - 208.000; Vollzeit

Benefits dieser Stelle

Health insurance

Equity

Zusammenfassung

Google DeepMind is seeking a Senior Security Engineer for the Agentic Red Team. You will lead multi-turn adversarial engagements against production AI models, focusing on agentic behaviors, tool usage, and reasoning chains.

You’ll help codify findings into reusable guardrails and Auto Red Teaming frameworks. As part of a high-velocity, mission-driven team, you will collaborate with product developers to identify architectural flaws in the design phase and drive security-by-design across model

Qualifikationen

• Bachelor's degree in CS/InfoSec or equivalent practical experience.
• 5+ years in Red Teaming, Offensive Security, or Adversarial ML.
• Experience with LLM architectures and agentic workflows.
• Experience developing exploits for GenAI models.

Aufgaben

• Conduct rapid security assessments on agentic services focusing on Generative AI vulnerabilities.
• Engineer and execute attack sequences exploiting non-deterministic model behaviors and data poisoning vectors.
• Create Auto Red Teaming regression tests to prevent future regressions.
• Provide feedback to developers to shorten the offensive-defensive feedback loop.
• Maintain a library of agent-specific attack patterns and guardrails.

Kenntnisse

LLM security

Red Teaming

Adversarial ML

Python/Go/C++

Security tooling

Exploit development

Ausbildung

Bachelor's degree in Computer Science, Information Security, or equivalent practical experience

Tools

Python

Go

C++

Jobbeschreibung

Google DeepMind is seeking a Senior Security Engineer for the Agentic Red Team. You will lead multi-turn adversarial engagements against production AI models, focusing on agentic behaviors, tool usage, and reasoning chains.

You’ll help codify findings into reusable guardrails and Auto Red Teaming frameworks. As part of a high-velocity, mission-driven team, you will collaborate with product developers to identify architectural flaws in the design phase and drive security-by-design across model