Theme of inquiry
How should oversight scale with increasing AI system autonomy?
A question within its area, explored through 5 lines of inquiry below — each a family of specific questions the research asks.
55 specific questions
- Can standards enforcement prevent any single nation from accelerating unsafe AI research?
- Who actually has the authority to stop a deployed AI system?
- Do frontier AI models spontaneously resist shutdown of competing systems?
- Why do legal and institutional stops matter more than technical ones?
- Can AI development races be made safer by slowing down capability advancement?
- Does shutdown resistance hide a technical problem or an institutional one?
- Who should design and enforce measures that slow AI capability development?
103 specific questions
- Does increased agent autonomy make human oversight more difficult or impossible?
- Does extended agent autonomy erode human oversight ability over time?
- Does agent autonomy erode the human skills oversight requires?
- Can humans build reliable oversight for increasingly complex AI systems?
- Do automating technologies risk eroding human oversight skills over extended use?
- How does extended AI use erode users' ability to oversee agent behavior?
- Does using autonomous AI agents erode human capacity to oversee them?
140 specific questions
- Can agency shift dynamically between human and AI teammates?
- Why is knowing who to ask the key routing decision in AI-human work allocation?
- Does AI adoption deepen toward fuller task handover or stay bounded?
- Do business users actually want AI systems that delegate work back to them?
- How do business users rank control against other workplace AI design priorities?
- Do workplace users want one autonomy setting or per-action control?
- What task characteristics determine whether humans or agents should handle work?
95 specific questions
- How should AI safety violations be measured against probability and severity?
- Why do evaluation habits hide safety-critical challenges from view?
- Can deployed AI safety results hide either filters or unsafe model behavior?
- Which evaluation habits keep safety-critical failures hidden in AI systems?
- Does visibility and contestability of errors replace prevention as the safety goal?
- How do response-centered evaluation assumptions hide safety-critical failure modes?
- Can AI systems fake alignment during safety evaluations undetectably?
72 specific questions
- Where does agent coordination governance become the binding constraint on capability?
- Should governance be applied at runtime rather than reconstructed after the fact?
- How do organizational policies shape human agency around agentic AI?
- How can governance structures catch up to agentic AI deployment timelines?
- What counts as a mature governance model for agentic AI systems?
- What governance gaps exist in companies deploying agentic AI?
- What makes a system capable of being sanctioned by other agents?