Tabletop Exercise Lead, The Anthropic Institute

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 Anthropic Institute

The Anthropic Institute (TAI) is a function within Anthropic that works to advance the public conversation around AI.

The Institute is comprised of research teams tackling some of AI's most consequential societal challenges: the Frontier Red Team, which stress tests our systems against catastrophic risks; Societal Impacts, which studies how Claude is used by millions of people; Economic Research, which tracks AI's impact on people's jobs and the larger economy; the Rule of Law team, which studies AI's impacts to democracy; and more.

The Institute's mission is to reduce information asymmetries between Anthropic and the outside world, name and answer the hard questions implied by powerful AI systems, and publish these findings so that policymakers, researchers, technology leaders, and the public can reckon with what's here and what's to come.

About the role

The role. You will build and lead the Institute’s tabletop exercise program and the team that runs it. Based on trends visible in the exponential and in consultation with technical experts, you will set the research agenda: which questions the exercises should answer, which scenarios to run, with which audiences, and what each run is meant to test. You will own the methods that make results comparable across runs and turn them into findings people outside the room can use. You will hire and manage a team plus a bench of facilitators, and you will be the program’s public voice.

Why tabletop exercises? If RSI arrives, the hardest problems will be about making decisions under uncertainty. The people making them may not trust each other, and the authorities they work under may not be fit-for-purpose. A tabletop exercise (TTX) allows the people who would face that situation practice it before it is real. It also shows us what tools and information they reach for, including what they would need that does not exist yet.

Tabletop exercises are also a research method. You can run the same scenario many times, with different groups of people (and with many more AI agents playing in silico), and you start to see the distribution of decisions people make. This is one of the few ways to generate empirical evidence about the future, i.e., how people and institutions will respond to situations that have not happened yet.

Key responsibilities

-

Set the program’s research agenda: the questions exercises should answer, th