An LLM app sends user prompts to a third-party model API, retrieves internal wiki chunks via embeddings, and may call a calendar tool. Where is sensitive employee data MOST likely to leave the corporate trust boundary first?
- AWhen the model’s tokenizer merges rare subwords
- BWhen the CSS theme is loaded from a CDN
- CWhen prompts and retrieved context are sent to the third-party model API
- DWhen the browser renders Markdown bold text