Search:
Match:
5 results
Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 08:53

A Comparative Study of Light-weight Language Models for PII Masking and their Deployment for Real Conversational Texts

Published:Dec 21, 2025 05:58
1 min read
ArXiv

Analysis

This article describes a research paper focusing on the application of lightweight language models for Personally Identifiable Information (PII) masking in conversational texts. The study likely compares different models in terms of their performance and efficiency for this specific task, and also explores the practical aspects of deploying these models in real-world scenarios.
Reference

“”

PermalinkArXiv
Software Development#Privacy & Security👥 CommunityAnalyzed: Jan 3, 2026 16:42

Local Privacy Firewall - Blocks PII and Secrets Before LLMs See Them

Published:Dec 9, 2025 16:10
1 min read
Hacker News

Analysis

This Hacker News article describes a Chrome extension designed to protect user privacy when interacting with large language models (LLMs) like ChatGPT and Claude. The extension acts as a local middleware, scrubbing Personally Identifiable Information (PII) and secrets from prompts before they are sent to the LLM. The solution uses a combination of regex and a local BERT model (via a Python FastAPI backend) for detection. The project is in early stages, with the developer seeking feedback on UX, detection quality, and the local-agent approach. The roadmap includes potentially moving the inference to the browser using WASM for improved performance and reduced friction.
Reference

“The Problem: I need the reasoning capabilities of cloud models (GPT/Claude/Gemini), but I can't trust myself not to accidentally leak PII or secrets.”

PermalinkHacker News
Research#llm🔬 ResearchAnalyzed: Jan 4, 2026 09:58

Randomized Masked Finetuning: An Efficient Way to Mitigate Memorization of PIIs in LLMs

Published:Dec 2, 2025 23:46
1 min read
ArXiv

Analysis

This article likely discusses a novel finetuning technique to address the problem of Large Language Models (LLMs) memorizing and potentially leaking Personally Identifiable Information (PIIs). The method, "Randomized Masked Finetuning," suggests a strategy to prevent the model from directly memorizing sensitive data during training. The efficiency claim implies the method is computationally less expensive than other mitigation techniques.
Reference

“”

PermalinkArXiv
AI Safety#LLM Security👥 CommunityAnalyzed: Jan 3, 2026 06:48

Credal.ai: Data Safety for Enterprise AI

Published:Jun 14, 2023 14:26
1 min read
Hacker News

Analysis

Credal.ai addresses enterprise concerns about data security when using LLMs. The core offering focuses on PII redaction, audit logging, and access controls for data from sources like Google Docs, Slack, and Confluence. The article highlights key challenges: controlling data access and ensuring visibility into data usage. The provided demo video and the focus on practical solutions suggest a product aimed at immediate enterprise needs.
Reference

“One big thing enterprises and businesses are worried about with LLMs is “what’s happening to my data”?”

PermalinkHacker News
Technology#AI Ethics/Privacy👥 CommunityAnalyzed: Jan 3, 2026 06:20

Ask HN: GPT-3 reveals my full name – can I do anything?

Published:Jun 26, 2022 12:37
1 min read
Hacker News

Analysis

The article discusses the privacy concerns arising from large language models like GPT-3 revealing personally identifiable information (PII). The author is concerned about their full name being revealed and the potential for other sensitive information to be memorized and exposed. They highlight the lack of recourse for individuals when this happens, contrasting it with the ability to request removal of information from search engines or social media. The author views this as a regression in privacy, especially in the context of GDPR.

Key Takeaways

Reference

“The author states, "If I had found my personal information on Google search results, or Facebook, I could ask the information to be removed, but GPT-3 seems to have no such support. Are we supposed to accept that large language models may reveal private information, with no recourse?"”

PermalinkHacker News