Security Engineer

About the role

Metr is building out a dedicated security function, and this position is among the first hires for that team. The organization handles pre-release frontier AI models, confidential lab data, and sensitive evaluation transcripts, making security foundational to maintaining trust with AI labs, governments, and the public. You will protect infrastructure that runs untrusted AI-generated code at scale while defending against both external attackers and internal agent-based threats.

Key facts

What you'll do

• Conduct targeted red-team exercises against internal systems and develop automated AI-driven red teaming capabilities
• Build AI-powered detection and response systems capable of triaging threats from both external attackers and internal agents
• Handle detection engineering, telemetry pipelines, incident response, and hardening across cloud infrastructure, endpoints, and identity systems
• Secure evaluation infrastructure that executes untrusted model-generated code during multi-day tasks
• Partner with researchers to make dangerous-capability experiments safe to run, addressing reward hacking and evaluation awareness scenarios
• Manage endpoints, development environments, cloud platforms, agent sandboxing, VPN and VPC networking, application code reviews, and access control

Requirements

• Strong foundational knowledge across systems, networks, cloud, and identity security
• Hands-on offensive security experience through red teaming, penetration testing, or adversarial research
• Ability to build with AI, including agent pipelines, LLM-powered tooling, and automated workflows, with understanding of current tool limitations
• Deep AWS knowledge, particularly around IAM policies

Nice to have

• Experience with SIEM and detection pipelines, writing and tuning detections, and threat hunting at scale
• Background in cloud and container security, including Kubernetes and infrastructure-as-code environments
• Track record leading or working severe incidents, especially those involving AI agents
• Familiarity with AI security research areas such as prompt injection, agent containment, model supply-chain risks, or red teaming AI systems

Skills & tools

• AWS services: EKS, Lambda, ECS, IAM, SQS, CloudWatch, SecurityHub, GuardDuty
• PostgreSQL with RLS and serverless Aurora
• Pulumi for infrastructure-as-code
• DataDog for SIEM
• Okta and Google Workspace for identity
• Tailscale for networking
• CrowdStrike Falcon for endpoint security

Practical notes

• Technical team members work from the Berkeley office 3-5 days per week, though flexibility is available
• Cap-exempt H-1B visa sponsorship is available for candidates who want to work in person but lack US work authorization
• No screening on certifications, degrees, or years of experience
• Applications encouraged even if your background does not appear to be a perfect match