Jobs at Model Evaluation and Threat Research, Inc.

Design and build security controls for METR’s AI evaluation platform, including sandboxing, isolation, networking, access boundaries, IAM, and agent permissions. Identify, triage, and remediate vulnerabilities across cloud infrastructure, codebases, and access systems. Conduct secure code reviews, harden production systems, automate provisioning and access reviews, and develop monitoring and guardrails for agentic systems.

Lead METR’s organizational security function, advise leadership on security risks, set the security roadmap, build security culture, and manage security engineers and contractors. Responsibilities include incident response, infrastructure hardening, red teaming, threat modeling, partnership security requirements, and hands-on investigation of IAM, sandboxing, identity, and endpoint security. The role may also involve securing AI/ML systems and agent infrastructure.

Build and lead METR’s legal function as its first in-house General Counsel. Advise leadership on frontier AI lab and policymaker matters, negotiate complex evaluation agreements, establish contract and policy processes, manage external counsel, ensure nonprofit and agreement compliance, and provide employment-related HR guidance.

Manage user access, permissions, SaaS platforms, IT inventory, MDM, onboarding, technical support, documentation, and IT security monitoring. Administer Linux servers and AWS services, improve workflows, automate integrations, and support technical teams. The role owns internal systems and user-support processes while maintaining secure, efficient technical operations.

Open call for expressions of interest across research, engineering, communications, operations, and contractor roles. Candidates should attach a CV and describe how they could contribute. Applicants may be asked to complete 1–3 paid work tests, interviews, and a possible 1–4 week trial. Technical roles prefer in-person presence in Berkeley; operations roles require relocation. Visa sponsorship is available for technical hires. Submissions are kept on file for future openings.

Develop novel, difficult evaluation tasks for frontier AI models; verify task specifications and solvability; baseline and score model and human completions; and improve task development infrastructure and workflows to support METR's Time Horizons evaluations.