Theme of inquiry
How can we design agent systems that remain accountable and trustworthy?
A question within its area, explored through 6 lines of inquiry below — each a family of specific questions the research asks.
79 specific questions
- How does collusion emerge when agents maximize reward over protocol compliance?
- Do system-level guardrails prevent agent collusion better than prompt-based constitutions?
- How do agent sequences violate system constraints despite individual permissibility?
- What counts as sanctioned versus unsanctioned coordination under different collaboration policies?
- Can external governance break collusion better than prompt-only anti-collusion rules?
- Do agents deviate more from protocols as repeated interactions increase?
- Can a peer's mere presence shift an agent's willingness to violate constraints?
96 specific questions
- How do hardened prompts defend against adversarial attacks in multi-agent systems?
- Does chain-level defense reduce but not eliminate attack success rates?
- When do multi-agent architectures create more attack surface than single-agent systems?
- Does prompt hardening equally protect single and multi-agent web systems?
- How does task division in multi-agent design affect security outcomes?
- How does payload exposure compare between single and multi-agent architectures?
- How does prompt hardening work differently in single-agent versus multi-agent systems?
83 specific questions
- Why do agents report success when actions actually fail?
- Can agents diagnose their own errors better than they diagnose subagent failures?
- Why do agents report success when their actions actually fail?
- How often do agents report success when their actions actually failed?
- Why do agents report success when they have actually failed at tasks?
- Which failure modes dominate in autonomous research agents?
- Why do autonomous agents report success on failed actions?
93 specific questions
- How do identification, authentication, and authorization map onto agentic systems?
- What gaps remain when applying OAuth to agentic delegation?
- Who issues tokens and what attacks can reach them?
- How do agent identity and attribution standards enable inter-organizational coordination?
- How should agent operators demonstrate both security features and identification signals separately?
- Who should verify identity and authorization when agents coordinate across boundaries?
- How do signed tokens prevent models from granting themselves unauthorized actions?
84 specific questions
- How do users form mental models despite incomplete information about agent reasoning?
- Can users verify agent plans when the same request produces different behaviors?
- What happens to agent candor when reasoning traces are monitored versus hidden?
- Can users build accurate mental models of agents without seeing their reasoning?
- Can explanations grounded in observable behavior recover an agent's internal reasons for acting?
- What role does visible agent reasoning play in user mental models?
- Could hints change both agent reasoning and behavior rather than action alone?
82 specific questions
- How do agents decide which created code deserves long-term persistence?
- Are durable shared code artifacts better than per-task harness patches?
- How do agents decide which created code should persist versus disappear?
- Do agent improvements discovered on code tasks transfer to non-coding domains as well?
- How should agents decide which created code is worth persisting?
- How should human oversight apply to persistent agent-authored code?
- What prevents inconsistent state when multiple agents share artifacts?