Foundation Model
A general-purpose AI model trained on broad data and adapted to many downstream tasks, including many LLMs and image models used in insurance applications.
A foundation model is a general-purpose AI model trained on broad data at scale and then adapted to many specific tasks. Most large language models and many image-generation models are foundation models. They are developed by a few large technology companies and then licensed or accessed through APIs.
In insurance, foundation models are typically used through third-party platforms. A carrier might use a foundation model to draft emails, summarize claims, or power a customer-service chatbot. The carrier usually does not own or fully understand the training data, which creates governance and vendor-risk questions.
Foundation models are powerful but opaque. A carrier using one should have clear contracts, understand data-use restrictions, and test outputs for accuracy and bias in the specific insurance use case. See our glossary entries on large language models, generative AI, and AI vendor risk assessment.