LLM gateway masks personal data and logs only the redaction, not the PII
The article describes how a self-hosted AI gateway called BaseLLM handles personal data pasted into prompts. Instead of keeping original prompts for audit, which would duplicate the sensitive data, the system scans text on the server before it reaches the model. Projects can choose from five modes, including report-only, masking, placeholder substitution, redaction, and blocking, while secrets like API keys are blocked by default. Detection uses rule-based patterns with check-digit validation to reduce false positives, though names and addresses are not covered. Only masked samples and counters are stored, and detection events are retained for 90 days by default.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in