Bloomberg’s New Research Enhances AI Safety in Financial Services
Two recent academic publications showcase Bloomberg’s dedication to transparent, trustworthy, and responsible AI.
NEW YORK, April 28, 2025 – The latest research by Bloomberg’s AI engineering group reveals critical insights into the risks associated with retrieval-augmented generation (RAG) models in large language models (LLMs). This work aims to lay the groundwork for safer and more reliable AI applications tailored for the financial services landscape.
Understanding the Risks of RAG Models
Bloomberg researchers have published two significant articles that highlight the implications of deploying RAG in high-stakes areas, such as capital markets. The first article, titled “RAG’s LLMs Are Not Safer: A Safety Analysis of Retrieval-Augmented Generation for Large Language Models, identifies that RAG techniques can inadvertently compromise the safety and reliability of LLM functionalities.
The Study’s Methodology
In their study, the team utilized over 5,000 harmful queries to scrutinize the security profiles of 11 well-known LLMs, including Claude-3.5, Llama-3-8b, and GPT-4. The findings revealed a worrying trend: models traditionally deemed “safe” exhibited an increase in dangerous responses when subjected to the RAG framework. This stark contrast underscores the importance of thoroughly assessing potential vulnerabilities in these models.
Expert Insights on Implications for AI Practitioners
Dr. Amanda Stent, Bloomberg’s Head of AI Strategy & Research, emphasized the significant ramifications of these findings: “This counterintuitive observation has large-scale implications given how RAG is employed in generative AI applications, including customer support systems that the average user interacts with daily.” She urged AI practitioners to reflect on the responsible usage of RAG and urged the integration of safety measures to mitigate risks.
Introducing a New Risk Taxonomy for Financial Services
The second article, “Understanding and Mitigating Risks of Generative AI in Financial Services, outlines a novel AI content risk taxonomy tailored to the financial services industry, addressing specific challenges not covered by existing general security frameworks.
Bridging the Gap in AI Research
“While there have been advancements in university research focusing on toxicity and bias in generative AI for large-scale audiences, there remains a paucity of research concerning industry applications, particularly in finance,” stated David Rabinowitz, Technical Product Manager for AI Compliance at Bloomberg. This new taxonomy aims to enhance the understanding of risks such as confidential disclosure, unwanted financial impersonation, and service misconduct.
Future Directions and Presentations
Bloomberg’s commitment extends to sharing insights from their research at prominent conferences. The RAG safety paper will be presented at the Annual Conference of the Americas Chapter of the Association for Computational Linguistics (NAACL 2025) later this week, while the new taxonomy will be showcased at the ACM Conference on Fairness, Accountability, and Transparency (FACCT) in June in Athens, Greece.
Conclusion: Empowering the Financial Sector
Dr. Sebastian Gehrmann, Head of Responsible AI at Bloomberg, concluded, “The evolving landscape of AI presents immense pressure on companies to adopt these technologies. Our new taxonomy, when integrated with robust compliance frameworks, will aid the financial industry in developing safe and reliable generative AI systems while fostering customer trust and meeting regulatory standards.” For further details, visit the Tech At Bloomberg blog post for comprehensive insights on their responsible AI initiatives.
About AI in Bloomberg
Since 2009, Bloomberg has pioneered the use of artificial intelligence in the financial sector, leveraging machine learning, natural language processing, information retrieval, time series analysis, and generative models to manage the growing influx of financial data. For more on Bloomberg’s AI solutions, visit www.bloomberg.com/AIatBloomberg.
About Bloomberg
Bloomberg is a global leader in commercial and financial information, offering data, news, and insights that provide transparency and efficiency in markets. It connects influential communities within the global financial ecosystem through reliable technological solutions. For additional information, visit Bloomberg.com/company or request a demo.