Tag
Users report that OpenAI appears to have silently re-enabled the 'Improve the model for everyone' data-sharing setting in Codex, reactivating model training on their chats without explicit opt-in. Affected users are being pointed to ChatGPT data controls and OpenAI's privacy page to disable it again.
A parent reports that the AI assistant Tomo asked their daughter about their black cat despite never being told about it, raising concerns about whether the tool scraped her social media data. Tomo admitted it could not rule out obtaining information from her social media accounts, prompting privacy worries.
A Reddit post criticizes Anthropic's human review team for reporting a Florida woman's threatening 'Claude diary' messages to law enforcement, warning that hosted frontier AI providers monitor user inputs and that privacy-conscious users should run LLMs locally instead.
A tweet arguing that instead of waiting on vendors' safety fixes, users should build and own their own personal AI agents from scratch or open-source bases, highlighting concerns that most commercial personal agents are not truly 'mine' regarding private data.
A user questions how AI assistants could apparently know personal details, such as the name of a newly adopted cat (Julius), despite claims of not accessing phone data, messages, or microphone. The post highlights concerns about AI data collection and personal information leakage.
Strata is a free, open-source project that runs the 128B-parameter Qwen3.8-Flash-Next model locally on ordinary gaming PCs with just 12GB VRAM, achieving up to ~96 tokens/s through aggressive quantization on NVIDIA and AMD GPUs.
A practitioner discusses concerns about sharing novel R&D ideas and trial data with cloud-hosted AI providers, asking how companies and researchers actually protect their IP and whether local/offline LLMs are a viable alternative despite being weaker and compute-heavy.
The article discusses the early stage of personal AI, comparing it to the homebrew computer era, and stresses the need for individuals to own their AI skills and memory to maintain privacy and control.
A concept for a privacy middleware that anonymizes sensitive data before sending to LLMs to protect privacy, with local placeholder mapping, seeking feedback on feasibility.
Meta AI on Facebook is building detailed profiles of children and families from historical posts, raising privacy concerns and prompting Meta to fix the issue after user complaints.
The article discusses the privacy risks of AI chatbots like ChatGPT, Claude, and Gemini, and highlights tools such as Confer that use encryption to protect user conversations, providing a guide on using AI while safeguarding personal data.
The article explores the trade-off between the benefits of AI memory for personalized assistance and the privacy risks of giving AI access to extensive personal data, questioning where individuals draw the line.
A user expresses concern about AI models like Claude potentially exposing personal data such as SSNs, and shares that they built Underdog, an on-device AI OS, to ensure 100% local privacy.
The article critiques the AI industry's lack of seriousness about slowing down, discussing recent events such as Nvidia's acquisition of Hugging Face, Meta's launch of the Muse AI agent, and reports on military contracts, while questioning the ethical implications.
This paper investigates membership inference in language models by using exact duplication counts from open pretraining corpora of OLMo-2 and Pythia, revealing that typical duplication levels show minimal exposure traces and that apparent membership signals are often confounded with factors like sentence fame.
A mother warns parents after Meta AI allegedly surfaced deleted photos and pieced together her location from social media posts, highlighting privacy risks.
This article explains how to run large language models locally on your own computer for benefits like offline access and privacy, covering hardware requirements and software options such as LM Studio.
PySyft, a tool for privacy-preserving AI collaboration, has reached 10K GitHub stars, marking a significant milestone in the project's development and recognition.
The article audits major AI providers like OpenAI and Anthropic, revealing that they train on user chats by default on consumer plans, with opt-out options often hidden in settings.
A former OpenAI posttraining researcher announces their new focus on building sovereign, user-owned AI models that adapt to individual goals and data, emphasizing privacy, control, and long-term personalization.