The Capital Markets Artificial Intelligence Glossary

A type of AI system trained on vast amounts of text data to understand and generate human language. Popular examples include Claude, ChatGPT, Gemini and Llama.

LLMs can answer questions, summarize information and assist with research, but they do not inherently understand capital markets data, business rules or workflows. To be useful in enterprise environments, they typically need access to additional context, controls and data sources.

A software system that can understand a goal, gather relevant information and complete tasks on a user's behalf. Unlike a traditional chatbot, which is primarily designed to answer questions, an AI agent can interact with data sources, applications and workflows to achieve an objective.

In capital markets, AI agents may be used to analyze trading activity, generate reports or investigate unusual patterns. Their effectiveness depends on the quality of the data, context, a query or set of commands, and controls available to them.

Generates outputs based on probabilities rather than fixed rules. This means the same input can sometimes produce different results, depending on how the system interprets the information available to it. Most modern AI models, including LLMs, are probabilistic by nature.

This flexibility is a strength, allowing AI to summarize information, identify patterns and generate new ideas. However, probabilistic outputs can also vary in accuracy and consistency.

A method that produces the same output every time it is given the same input. The result is predictable, repeatable and consistent, making deterministic systems well suited to calculations, rules-based processes, and financial controls.

Examples in capital markets include benchmark calculations, risk models, compliance checks, and order validation rules. Unlike many AI models, deterministic systems do not "guess" or generate different answers to the same question.

When a model generates information that appears plausible but is factually incorrect, misleading, or unsupported by reliable evidence. Hallucinations can range from minor inaccuracies to entirely fabricated figures, sources or events.

Because many AI systems are probabilistic, hallucinations are a known limitation rather than a software error in the traditional sense. The risk can be reduced by grounding AI in trusted data sources, applying governance controls, and providing users with evidence to support outputs.

An AI technique that combines an LLM with information retrieved from trusted data sources. Instead of relying only on its training data, the model can access relevant documents, databases or knowledge repositories before generating a response.

This approach helps ensure answers are based on current, organization-specific information rather than general knowledge alone.

An open standard that allows AI models to securely connect with external data sources, applications and business systems. Rather than relying solely on the information they were trained on, AI tools can use MCP to access relevant data and perform tasks using approved resources.

For example, MCP can enable an AI assistant to retrieve information from trading platforms, databases or internal applications while operating within defined permissions and controls.

The ability to track where data originated, how it has moved through a system, and any transformations applied along the way. It provides a clear record of how a particular figure, report or AI-generated insight was produced.

In complex organizations, data often passes through multiple systems before reaching end users. Data lineage helps ensure transparency and makes it easier to identify errors, validate outputs and support regulatory requirements.

The principle that an organization retains control over how and where its data is stored, processed, and governed. Data sovereignty requirements are often shaped by local regulations, security policies and contractual obligations, particularly when data is held in cloud environments or used by third-party providers.

For financial institutions, maintaining control over sensitive client, trading, and operational data is a critical consideration when evaluating AI solutions and technology platforms.

A structured framework that defines the concepts, entities and relationships within a domain. In capital markets, an ontology establishes what terms such as orders, positions, portfolios and benchmarks mean, and how they relate to one another. It creates a shared understanding of business concepts that can be understood consistently by both humans and machines.

The practical framework that applies agreed business definitions to data across an organization. It acts as a common language between systems, allowing users and applications to access information using consistent terms and metrics, regardless of where the underlying data resides.

The ability to understand how an AI system arrived at a particular output, recommendation or decision. An explainable system provides visibility into the data, logic and processes that contributed to a result, helping users assess its accuracy and relevance.

In financial services, where decisions often need to be justified to clients, regulators and internal stakeholders, explainability is a key component of trustworthy AI.

An approach to AI governance that keeps people involved in reviewing, approving or refining AI-generated outputs before any actions are taken. Rather than operating entirely autonomously, the AI supports human decision-making by providing recommendations, insights or analyses.

In financial services, human oversight is particularly important when AI is used in areas such as trading, risk management and compliance, where errors can have significant consequences.

A chronological record of actions, decisions, and system activity that can be reviewed and verified. In AI systems, an audit trail may capture the data used, prompts submitted, outputs generated and actions taken in response to those outputs.

In financial services, audit trails play an important role in supporting transparency, governance, and regulatory compliance. They can help firms understand how a decision was reached and provide evidence if that decision is later questioned.

The policies, controls and oversight processes that organizations put in place to manage the responsible use of artificial intelligence and its outputs. It covers areas such as data quality, security, compliance, explainability, accountability and human oversight.

In financial services, AI governance helps ensure that systems operate in line with regulatory requirements, internal policies, and risk management frameworks.