Security Engineer, DeepMind Agentic Detection and Response

Our Security team works to create and maintain the safest operating environment for Google's users and developers. Security Engineers work with network equipment and actively monitor our systems for attacks and intrusions. In this role, you will also work with software engineers to proactively identify and fix security flaws and vulnerabilities.

As a Security Engineer on DeepMind Agentic Detection and Response in New York, you will be at the intersection of security operations, trajectory forensics, and AI-powered detection engineering. Because autonomous agents operate at machine speed, we practice "agent-first" defense: our automated pipelines and multi-agent workflows absorb Tier-1 volume and trigger containments, freeing our engineers to manage high-ambiguity investigations, hunt for novel agentic behaviors, and build the detections and agentic tooling that keep pace with rapidly evolving frontier models.

In this role, you will have a true operational and engineering split in a fast-moving, 0-to-1 environment where models, agent harnesses, sandboxes, and telemetry evolve weekly. Because operational security is core to safeguarding continuous frontier training and evaluation runs, this position includes rotational on-call support, including weekday daytime triage and periodic weekend on-call coverage. You might reconstruct multi-step Chain-of-Thought and tool-call trajectories to determine whether an anomaly is benign specification gaming or a real sandbox escape, coordinate live containment and blast-radius scoping between DeepMind teams and PSS Detection and Response, and build new AI agent skills, detection rubrics, or automated forensic pipelines in Python and SQL. You will deliver substantial technical and investigative contributions across interconnected components with minimal assistance, owning scoped end-to-end investigations, detections, and engineering artifacts, and codifying every catch into durable guardrails.

Individual pay is determined by factors including job-related skills, experience, and relevant education or training.

US: $147000 - $210000 (USD) + 15% bonus target + equity + benefits

Learn more about benefits at Google.

Responsibilities

• Reconstruct multi-step Chain-of-Thought (CoT) and tool-call trajectories to determine if anomalies are benign specification gaming or sandbox escapes.
• Coordinate live containment and blast-radius scoping between DeepMind teams and PSS Detection and Response.
• Build new AI agent skills, detection rubrics, and automated forensic pipelines using Python and SQL.
• Investigate high-ambiguity security events and hunt for novel agentic behaviors across frontier models.
• Develop detections and agentic security tooling to keep pace with rapidly evolving frontier models.

Minimum qualifications:

• Bachelor's degree or equivalent practical experience.
• 2 years of experience with security engineering, computer and network security and security protocols.
• 2 years of experience with security assessments or security design reviews or threat modeling.
• 2 years of coding experience in one or more general purpose languages.
• Experience utilizing Agentic AI frameworks or developer kits to build and deploy autonomous AI systems.
• Experience in a technical security role (e.g., security engineering, security research, DevSecOps, incident response, SOC analyst, incident response, malware research, threat analysis).

Preferred qualifications:

• 5 years of experience in security engineering, detection engineering, incident response, or threat hunting.
• Experience investigating security events across Linux, containerized or sandboxed environments, cloud infrastructure, or corporate endpoints.
• Experience building or tuning behavioral detections, LLM-based evaluators, or multi-agent workflows, and measuring detection effectiveness.
• Experience operating in an on-call capacity, managing security escalations, and working with cross-functional teams.
• Experience with internal security, data, or AI ecosystems.
• Knowledge of frontier AI and agentic security risks, sandbox escape, prompt injection, or training environment tampering.