<![CDATA[#OpenAI releases #PrivacyFilter โ€” an open-weight #AI model for detecting &amp; redacting #PII in text.]]> releases โ€” an open-weight model for detecting & redacting in text. Runs fully locally, no data ever leaves your machine. Apache 2.0 licensed.

๐Ÿงต๐Ÿ‘‡

๐Ÿ” Detects 8 PII categories in a single forward pass: names, email addresses, phone numbers, physical addresses, URLs, dates, account numbers & secrets (passwords, API keys) โ€” covering virtually all common sensitive data types

Link Preview Image
]]>https://board.circlewithadot.net/topic/dd45c278-ea2c-4c8e-b7ae-29c3e4eb28d9/openai-releases-privacyfilter-an-open-weight-ai-model-for-detecting-amp-redacting-pii-in-text.RSS for NodeFri, 15 May 2026 00:23:49 GMTSat, 25 Apr 2026 00:43:46 GMT60<![CDATA[Reply to #OpenAI releases #PrivacyFilter โ€” an open-weight #AI model for detecting &amp; redacting #PII in text. on Sat, 25 Apr 2026 02:17:43 GMT]]>@michabbb Finally, a use for local LLMs.

]]>https://board.circlewithadot.net/post/https://social.vir.group/users/newsgroup/statuses/116462941836603708https://board.circlewithadot.net/post/https://social.vir.group/users/newsgroup/statuses/116462941836603708Sat, 25 Apr 2026 02:17:43 GMT<![CDATA[Reply to #OpenAI releases #PrivacyFilter โ€” an open-weight #AI model for detecting &amp; redacting #PII in text. on Sat, 25 Apr 2026 00:43:47 GMT]]>๐Ÿ”ง Fine-tunable on domain-specific data โ€” adapts to medical, legal or enterprise environments where generic rules fail. Based on the open model family. Available on under Apache 2.0

๐Ÿšจ Caveat: is a redaction & data minimization aid โ€” NOT a compliance guarantee. It should be one layer in a holistic approach. Always combine with human review for high-stakes use cases
https://openai.com/index/introducing-openai-privacy-filter/

]]>https://board.circlewithadot.net/post/https://social.vivaldi.net/users/michabbb/statuses/116462572516011291https://board.circlewithadot.net/post/https://social.vivaldi.net/users/michabbb/statuses/116462572516011291Sat, 25 Apr 2026 00:43:47 GMT<![CDATA[Reply to #OpenAI releases #PrivacyFilter โ€” an open-weight #AI model for detecting &amp; redacting #PII in text. on Sat, 25 Apr 2026 00:43:47 GMT]]>๐Ÿ“ 128,000-token context window โ€” processes entire legal documents, long email threads or large codebases in a single pass. No need to chunk text before filtering.

๐Ÿ› ๏ธ Built for high-throughput workflows: CLI tool (opf), GPU & CPU support, interactive mode, structured JSON output with ANSI color-coded previews. Runs on-premises โ€” data never sent to external servers

]]>https://board.circlewithadot.net/post/https://social.vivaldi.net/users/michabbb/statuses/116462572474306873https://board.circlewithadot.net/post/https://social.vivaldi.net/users/michabbb/statuses/116462572474306873Sat, 25 Apr 2026 00:43:47 GMT<![CDATA[Reply to #OpenAI releases #PrivacyFilter โ€” an open-weight #AI model for detecting &amp; redacting #PII in text. on Sat, 25 Apr 2026 00:43:46 GMT]]>๐Ÿง  Bidirectional token-classification โ€” unlike autoregressive LLMs, reads input from both directions simultaneously for deeper context awareness, catching subtle that simple pattern-matching or RegEx rules miss

โšก 1.5B parameter model with only ~50M active parameters () โ€” lightweight enough to run on a standard laptop or in a browser, yet achieves ~96โ€“97% F1 score on standard benchmarks

]]>https://board.circlewithadot.net/post/https://social.vivaldi.net/users/michabbb/statuses/116462572442988099https://board.circlewithadot.net/post/https://social.vivaldi.net/users/michabbb/statuses/116462572442988099Sat, 25 Apr 2026 00:43:46 GMT