Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Aug 13·5 min read
Governing Agentic AI in FinTech
TL;DR
Agentic AI systems in FinTech are increasingly making significant decisions with minimal oversight, raising concerns about their verifiability. This study introduces the Verifiability Gap, highlighting the need for robust governance frameworks to ensure accountability in AI decision-making.
✦ Why It Matters
Evaluate your AI systems' verifiability today to ensure compliance and accountability in decision-making.
Key Takeaways
- Assess the verifiability of your AI systems against regulatory standards.
- Implement controls to enhance reproducibility in AI decision-making.
- Understand the limitations of commercial AI models in auditability.
- Adopt a governance framework that prioritizes evidence-based delegation.
How It Works
The study develops a multilevel governance theory for agentic AI, focusing on the Verifiability Gap. It tests various AI models, analyzing how changes in architecture and controls affect decision reproducibility.
The findings indicate that local models can consistently reproduce decisions under tight controls, while commercial models exhibit variability, suggesting that capability does not equate to auditability.