Third Annual Risk Conference 2025
What it is about
Omar Bairán, CFA, FRM and CQF, explained at the Third Annual Risk Conference (September 18, 2025) how artificial intelligence (AI) is regulated and what risks its adoption poses in financial services. The opinions are solely the speaker's.
The 41 slides compare European Union regulation (by risk tiers) with the United States approach, review cases of failed implementations, the risks of generative AI and AI agents, an example of model risk with a synthetic credit portfolio, algorithmic discrimination, AI governance frameworks and explainable AI methods.
Objectives
- Understand the balance between protection and innovation in the regulatory design of AI.
- Identify why many AI projects fail and what new risks AI introduces in the financial sector.
- Present the governance frameworks and explainable AI techniques to manage those risks.
Key takeaways
- AI regulation must balance the protection of fundamental rights, security and accountability against the risk of slowing innovation, raising research costs and quickly becoming obsolete.
- The European Union AI Act classifies uses by risk level: unacceptable ones are prohibited and high-risk ones require registration and conformity assessment; the United States relies on a mix of executive orders, sector laws and agency guidance.
- An MIT report cited indicates that 95% of generative AI pilots in financial institutions produced no measurable savings or benefits; the most common causes are poor data (35%), organizational resistance (28%) and poor business fit.
- An example with a synthetic credit portfolio shows that optimizing only the area under the curve can leave the probability of default poorly calibrated and distort expected loss.
- AI governance relies on frameworks such as NIST AI RMF, ISO/IEC 42001, the EU AI Act and the OECD principles, and explainable AI (LIME, SHAP or interpretable models) responds to the regulatory trend toward greater transparency.
Connect with our networks