pii

"OpenAI recently released a model for detecting personal information called Privacy Filter. It is lightweight at 1.5B and is said to work even on a laptop's CPU."

"By running inference entirely within the user's browser (locally), there is no need to send confidential data to external servers, thereby minimizing the risk of information leakage to the absolute limit."

"New model for detecting and masking PII from OpenAI"

"The official position is "frontier-level personal data detection in a small model that can be executed locally"."

"OpenAI Privacy Filter is a bidirectional token classification model designed for the detection and masking of PII (personally identifiable information), capable of running in on-premises environments."

"You've stripped PII from prompts before they reach the model. You have audit logs proving it. And yet - the logs might still contain PII, just through a side door you didn't think to close."

"OpenAI announced the release of 'Privacy Filter', an open-source lightweight model for masking personal information in text, designed to remove personal data before it is read into AI chatbots, available for commercial customization."

"OpenAI releases OpenAI Privacy Filter as open source to prevent leakage by masking personal information such as personal names, email addresses, and phone numbers in documents."

"OpenAI releases Privacy Filter, an open-weight model for masking personally identifiable information in text, with 1.5B total and 50M active 参数"

"OpenAI Privacy Filter is an open-weight model for detecting and redacting personally identifiable information (PII) in text with state-of-the-art accuracy"

"Personally, if you organize Guardrails by 'who makes the final decision,' it becomes much easier to understand all at once. If the app decides, it's On-Demand; if you want OCI to forcibly stop it, it's Block; and if you want to observe and audit first, it's Inform."

"The problem has two layers. The first layer is the risk of sending data to external LLM providers... The second layer is the risk of recording it in your own company's logs."

"CloakLLM is open-source middleware that detects PII, replaces it with reversible tokens, and restores originals in the response."

"CloakLLM is open-source middleware that detects PII, replaces it with reversible tokens, and restores originals in the response."

"We use LLMs on Databricks to automatically detect and classify sensitive data across logs and databases."

"We ended up building our own observability and governance tool called Syntropy to handle this. It basically logs all the standard trace data (tokens, latency, cost) but focuses heavily on real-time guardrails—so it auto-redacts PII and blocks prompt injections before they execute, without adding proxy latency."

"AI Guardrails provides the ability to detect harmful content, defend against prompt injection attacks, and detect Personally Identifiable Information (PII) through the ApplyGuardrails API."

"This server is designed to allow AI (currently mainly Claude Desktop, etc.) to determine whether Personally Identifiable Information (PII) is included in the text."