The AI in Business Podcast How Financial Services Are Building Safer Customer-Facing AI - with Akhil Khunger of Barclays
20 snips
May 12, 2025 Akhil Khunger, VP in Quantitative Analytics at Barclays, shares insights on customer-facing AI in financial services. He discusses the evolving safety measures being integrated into these systems, tackling risks like prompt injection and model jailbreaking. Akhil highlights the balancing act between enhancing user experience and adhering to regulatory requirements. He further explores the challenges of building transparent and trustworthy AI systems, emphasizing the significance of user feedback and robust governance frameworks.
AI Snips
Chapters
Transcript
Episode notes
Build Strong AI Input Guardrails
- Implement layers to filter out harmful inputs like viruses before data reaches AI models.
- Build rigorous test cases and ensure models produce errors on failures to protect customers from unsafe outputs.
Risks From Data and Prompt Attacks
- Flawed training data and malicious prompt inputs pose substantial risks to financial AI systems.
- Special attention is required to prevent bugs or unauthorized actions triggered by specific commands.
Use Dashboards for Safety Monitoring
- Create safety dashboards with clear metrics, warnings, and comprehensive system monitoring.
- Monitor impacts across multiple accounts to detect indirect risks from AI prompt use.