Anthropic wants crisis exercises that can be repeated and compared
Anthropic’s public research agenda already proposes tabletop fire drills. Its Institute vacancy adds standardized scenarios and data capture, repeated runs with people and AI agents, and a facilitator kit for other teams and partners. That could make differences in decisions and missing information easier to study. Simulated choices do not establish how real institutions will behave.
The constraint
Can a crisis rehearsal reveal more than what one group did in one room?
Another explanation
This may be standard exercise design applied to an existing agenda. Repeated runs could mainly teach participants, and AI agents may poorly represent institutions.
Evidence
- Anthropic · Tabletop Exercise Lead, The Anthropic Institute
standardize scenarios and data capture so results are comparable, and use repeated play to map the range of outcomes
Context
- The Anthropic Institute research agenda
The page displays May 7, 2026 and already asks how tabletop fire drills could test decisions by lab leadership, boards and governments. Current inspection does not independently verify historical availability.
What would change this reading?
A reusable protocol stating what stays constant, how participants are selected, and what comparisons reveal beyond a single exercise.
For candidates
Experimental design and comparable decision records are central to the remit.