How to apply for Research Engineer / Research Scientist, Remote Compute Accounting

About SASH

SASH is a Singapore-based nonprofit that builds AI safety research capacity and governance expertise across Asia. Its Verification team prototypes tools to verify agreements about AI, works with policymakers, and partners with groups like the Future of Life Institute and University of Oxford. The technical team is led by Pascal Berrang, an Associate Professor at University of Birmingham, and includes experience from the Singapore Government.

About the role

This role focuses on software-only approaches to detecting undeclared training or computation inside AI datacenters, where a cluster could serve declared inference while training an undeclared model on leftover capacity. You would build and test verification mechanisms, contribute to international research collaborations, and help show policymakers what these tools can do. The work sits at the intersection of technical research and AI governance.

A typical day

A typical day might involve prototyping a verification mechanism, reviewing cluster logs or scheduling data, and discussing results with the technical lead and policy team. You could also join calls with international partners or prepare findings for policymakers. The team is small and remote, so expect frequent written communication and self-directed work.

Who SASH is looking for

• Has experience with distributed systems, cluster scheduling, or resource accounting in datacenter environments.
• Can design and prototype software-only verification mechanisms, not just analyze existing systems.
• Is comfortable working with both technical teams and policy-focused collaborators.
• Has a research background or engineering track record in systems, security, or AI infrastructure.

Tips for this application

• Read SASH's published work on verification and reference specific mechanisms or papers in your application.
• Explain how your background relates to detecting undeclared computation or resource accounting, even if your experience is in adjacent areas.
• Mention any experience working with policymakers, international collaborators, or cross-border technical projects.
• Note that the role is remote and based at a Singapore nonprofit; clarify your time zone and availability for collaboration across Asia and Europe.
• Avoid generic AI safety statements; focus on the specific technical problem of verifying what runs inside a datacenter.

What to cover in your cover letter

['Your technical approach to software-only verification of compute usage in shared clusters.', 'Any experience with distributed training, inference serving, or datacenter resource management.', 'How you have worked across technical and policy boundaries or in international research collaborations.', 'Why you want to work on verification for AI agreements specifically, and why at a Singapore-based nonprofit.']

Research before applying

• SASH's Verification team publications and prototypes, especially any work on compute accounting or datacenter verification.
• Pascal Berrang's research at University of Birmingham and his work with SASH.
• SASH's partners, including the Future of Life Institute and University of Oxford, and their roles in AI verification.
• Singapore's approach to AI governance and why a Singapore-based nonprofit is working on verification.

Likely interview topics

Based on the job description, expect questions about:

• How would you design a software-only system to detect undeclared training on a cluster that is also serving declared inference?
• What are the limits of software-only verification compared to hardware-based approaches?
• How would you test whether a verification mechanism works in practice?
• Describe a time you worked with policymakers or non-technical partners on a technical project.
• What do you know about SASH's Verification team and its current research directions?

Common mistakes to avoid

• Treating the role as a general AI safety position rather than a technical research engineering role focused on verification.
• Ignoring the policy context; this role requires working with policymakers and international collaborators.
• Assuming the problem is solved by hardware alone; the job description specifically asks for software-only approaches.