LSEG’s AI-Powered Market Abuse Detection on Amazon Bedrock
The London Stock Exchange Group (LSEG) leverages Amazon Bedrock and Anthropic’s Claude foundation model to enhance its market surveillance capabilities. Their AI-powered Surveillance Guide tackles the challenges of manually reviewing trade alerts for potential market abuse by automating the analysis of news sensitivity. The system ingests news articles from LSEG’s Regulatory News Service (RNS), preprocesses the data, and uses Claude Sonnet 3.5 via Amazon Bedrock to classify articles based on their potential market impact. This three-component architecture (data ingestion, Bedrock integration, and results visualization) processes articles, extracts clean text, prompts the model, and presents predictions via a Streamlit interface. A comprehensive dataset of 250,000 RNS articles was used for training and evaluation, with analysts providing price sensitivity labels. The system uses a two-step classification process to maximize accuracy, focusing on precision for non-sensitive news and recall for price-sensitive news. Key benefits include 100% precision in identifying non-sensitive news and 100% recall in detecting price-sensitive content during testing. This leads to reduced manual review, improved consistency, detailed audit trails, faster responses to potential abuse, and scalable surveillance. LSEG plans to expand the solution by integrating additional data sources, implementing few-shot prompting, and deploying it in live environments. The use of Amazon Bedrock provides secure, scalable access to foundation models, minimizing direct model management and enabling dynamic scaling.
LSEG’s implementation showcases how ai automation detection systems can revolutionize financial compliance by identifying suspicious trading patterns in real-time.
LSEG’s advanced system complements existing chatgpt automation detection tools by identifying sophisticated market manipulation patterns that traditional methods might miss.
