According to aws.amazon.com, Amazon Web Services has released a model-agnostic personally identifiable information (PII) detection system built on large language models. The detector, available as the “pii-detector” package in the sample-llm-pii-detection repository, runs on any LLM managed on Amazon Bedrock.
The system was evaluated on five public PII corpora across nine LLM-based detectors, including the OpenAI PrivacyFilter, according to aws.amazon.com. The detector is described as “configurable” and “instruction-driven.”
According to the AWS blog post, traditional PII detection tools use “bi-directional token-classification models: transformer taggers that label each token with a PII type fixed at training time.” The LLM-based approach addresses a key limitation: domain-specific identifiers like employee IDs or crypto-wallet addresses fall outside fixed schemas and typically require relabeling and retraining.
The new system reframes the problem by reading instructions at inference time, making “the entities to detect, the output format, and the deployment backend all become configuration rather than code,” according to aws.amazon.com. The detector can target new entity types by editing a prompt instead of retraining, and can “reason about context across eight languages without a translation step.”
The blog post notes that PII often appears in unstructured text like customer-support transcripts, HR records, and chat logs used in fine-tuning datasets.