About Anthropic
Anthropic’s mission is to create reliable, interpretable, and steerable AI systems. We want AI to be safe and beneficial for our users and for society as a whole. Our team is a quickly growing group of committed researchers, engineers, policy experts, and business leaders working together to build beneficial AI systems.
About the role
Anthropic gives independent external evaluators access to our models, our people, and our safety practices so that they can assess risks, processes, and policies. We're hiring a Technical Program Manager to establish and deepen our partnerships with these external reviewers and embedded evaluators to support a thriving frontier evaluation ecosystem.
You'll be the main point of contact for our existing partners, deepening our support so that they can work effectively and produce high-quality, rigorous, independent publications. You'll separately broaden the portfolio of organizations we are working with, managing the funnel from first conversation through contracting, onboarding, and day-to-day access, culminating in published reports. You'll evolve our current time-intensive manual processes into a system that can support many evaluators at once, working with teams across Safety, Research, Security, Legal, People Services, IT, Policy, and Communications. Within six months, you should be able to take a new evaluator from contract negotiation to first published report with minimal involvement from senior leaders.
You'll need the technical depth to work out with researchers and engineers what access, tooling, and evidence an evaluation requires, and the judgment to resolve most questions quickly and escalate the few that need it, while balancing evaluators' need for access with Anthropic's requirements to protect its customers, its security, and its confidential information. We're looking for a generalist who can ship quickly and iteratively, and who earns trust on both sides.
Key responsibilities
Own the external evaluator program end to end, including our relationships with evaluator organizations and the pipeline of future partners
Take each evaluator from first conversation through scoping, contracting, onboarding, access, and review to a published report
Define the stages and tiers of working with evaluators, and develop information access policies that meet evaluators' needs while protecting security and confidential information
Drive evaluators' access and information requests to fast, well-documented decisions in partnership with Research, Security, Legal, and IT
Drive the technical design of evaluator access with Security, IT, and infrastructure teams, including permissions, environments, and monitoring, and close gaps between policy and technical controls without slowing evaluators down
Run the review process for evaluator reports against agreed timelines and redaction criteria, while protecting evaluators' ability to publish their own conclusions
Scope evaluations with researchers and engineers, turning evaluators' questions about training, monitoring, and alignment into concrete requirements for model access, data, tooling, and compute, and planning capacity for each new evaluator
Turn evaluators' technical findings into research and engineering work with owners and dates, and drive accountability for how Anthropic responds
Keep the company informed with regular updates on how the program is going, and clear guidance on how to work with evaluators
Collect feedback from evaluators and internal teams, run retrospectives, and iterate on policies and practices
Use Claude and work with engineering teams to build the tooling the program runs on, such as access provisioning, request tracking, and report review
Support a thriving ecosystem of evaluators by sharing what we learn about running independent evaluation well
Minimum qualifications
Experience leading complex technical programs with engineering or research teams, from ambiguous goals through delivery
Experience managing…