M

METR

Independent evaluation of frontier AI models

Member of Technical Staff - Embedded Assessments

AI ResearchConfirmed live in the last 24 hours
Full-timeBerkeley, CA, USA$402k - $687.8k /yrHybrid

About the Job

Requirements

  • Candidates should have at least two of the listed skills: proficiency prompting large language models; familiarity with relevant research or ability to get up to speed quickly; some security experience; ability to get up to speed on large codebases quickly; ability to perform loss-of-control threat modeling and break down safety cases; strong verbal communication, writing, and stakeholder management, including navigating complex relationships with frontier labs; and trustworthiness with a good reputation.
  • Technical team members are expected to work in the Berkeley office 3–5 days per week; applicants should disclose if this is a constraint.
  • Applicants who lack U.S. work authorization and want to work in person may be eligible for cap-exempt H-1B sponsorship.

Responsibilities

  • Conduct embedded exercises at frontier AI labs, including investigating incidents, stress-testing internal agent monitoring systems, and assessing loss-of-control risks from internal deployment.
  • Between exercises, practice and develop the general methodology, talk to other researchers, and build tooling to improve future exercises.
  • Help hire and scale the team, write up results, and plan and coordinate future exercises.
  • Conduct assessments across AI research and development acceleration, compute allocation, monitorability red-teaming, incident investigations, and other developing exercises.

About the Company

METR

Independent evaluation of frontier AI models

METR evaluates frontier AI models for dangerous capabilities before deployment by partnering with leading AI companies to gain access to models like GPT-4 and Claude. Its approach uses red-teaming and capability evaluations to test for long-horizon, agentic tasks such as autonomous replication, rapid R&D, or cyberattacks, without accepting compensation. METR informs risk assessment methods and safety policies for developers and policymakers, contributing to frameworks like OpenAI's Preparedness Framework and Anthropic's Responsible Scaling Policy. Its goal is to reduce the risk of harmful AI capabilities being released and to strengthen AI governance through independent technical evaluation.

Company size
51-200
Company stage
Grant
Total funding
$71M
Headquarters
Berkeley, CA
Founded
2022

Recently Posted Jobs

Explore more opportunities at METR

See more jobs