Sabrina Shih

Sabrina works to bolster international cooperation on AI verification, with a focus on bridging Chinese and Western research communities. She is also a Research Affiliate at the Oxford Martin AI Governance Initiative and the Hardware AI Governance Lab. She previously conducted technical AI governance research as a Summer Fellow at the Centre for the Governance of AI. Her experience also includes consulting Fortune 500 companies on enterprise AI risk management and guiding policy and solutions for AI assurance within the U.S. context.

Projects

Research Direction: Technical and policy options for verifying U.S.–China AI agreements

International agreements could help reduce catastrophic AI risks, but their credibility depends on whether participants can verify that commitments are being upheld. What technical and political factors enable or limit possibilities for AI verification between the U.S. and China? This project will investigate how verification could work in China’s institutional and industry context, and what would make proposed approaches feasible and credible to both sides.

Depending on your interests and background, you could pursue various directions, e.g.,:

• Institutional pathways: Explore which types of organizations would need to participate in a verification arrangement, what responsibilities they could hold, and how coordination could work.
• Industry participation: Investigate how cloud providers, data center operators, and other technology companies could support verification, including practical requirements, commercial incentives, and protection of sensitive information.

Day to day, you would conduct Chinese- and/or English-source research, analyze policy and industry practices, speak with relevant experts where feasible, and develop recommendations through regular discussion and feedback. The emphasis is on connecting technical possibilities to workable governance arrangements.

By the end of the fellowship, a strong result would be a focused report or set of memos that assesses a concrete verification pathway, makes assumptions and uncertainties explicit, and recommends practical next steps for researchers, policymakers, or potential implementation partners.

This work could help make commitments addressing dangerous AI development and deployment more credible, supporting cooperation even where trust is limited. Your project is flexible, and we can choose and scope a question together based on your strengths, interest, and expected impact for SAIF"s verification workstream.

What I'm looking for in a Mentee

I’m looking for a fellow with familiarity with China’s technology industry or policy landscape and, if possible, strong professional Chinese reading and speaking skills and the ability to conduct careful Chinese-source research. Knowledge of cloud computing, data centers, or compute governance—and relevant industry or policy relationships—would be particularly valuable. You should be able to turn an open-ended question into a focused research project, engage with technical details, and write clear, actionable analysis for specific audiences. I’m open to different career stages and work best with someone who takes initiative, shares work early, makes uncertainties explicit, and welcomes collaborative feedback.

What I'm Like as a Mentor

I tend to take a collaborative approach to mentoring, directing you to impactful and relevant research projects based on your expertise. I like meetings with a clear and prepared agenda and like to spend as much time as needed to clarify ideas, provide context, and strategize. I am happy to arrange quick calls to unblock things and generally prefer calls for complex discussions versus asynchronous messages. I welcome questions of any size and encourage you to challenge my ideas.