Definition
The process of taking a general-purpose AI model and further training it on a specific organisation's own data — teaching it the company's terminology, style, policies, or specialised knowledge. The result is a customised model that performs better for that organisation's specific use cases.
Why it matters
Fine-tuned models contain a company's proprietary data baked into their weights. If an attacker gains access to a fine-tuning endpoint or the resulting model, they may be able to extract that proprietary information or introduce hidden malicious behaviours during the training process.