Model-agnostic PII detection with LLMs
A configurable, model-agnostic detector that turns any large language model on Amazon Bedrock into a PII detector. Because the entities to detect live in a prompt rather than in code, one detector adapts to new entity types without retraining, and it outperforms an off-the-shelf tool across five public corpora and nine LLM-based detectors.
The article discusses the development of a model-agnostic PII detection tool using large language models (LLMs) on Amazon Bedrock. The tool, called pii-detector, is configurable and can detect personally identifiable information (PII) across various LLM-based detectors. It addresses the issue of PII memorization in models trained on uncleaned text, which can lead to leakage of real person's details through prompts.
The detector treats the LLM as a configurable, swappable component and wraps the input text in instructions that define the PII entities to detect and the expected output. This allows for easy addition of new entity types by editing prompts instead of retraining. The tool is available as the pii-detector package and provides a uniform inference interface for reaching the language model.
Brief written by urgent.news from AWS Machine Learning's own syndicated text. Machine-written — may contain errors; check the original before relying on it.
