Google hired NCC Group to evaluate its Private AI Compute system, which aims to enhance mobile AI capabilities using cloud resources while maintaining user privacy. The review included two phases: an architecture assessment and a detailed security analysis of various components, involving ten consultants over 100 person-days.

This article investigates the data sent by seven popular AI coding agents during standard programming tasks. By intercepting their network traffic, the research highlights privacy and security concerns, revealing how these tools interact with user data and potential telemetry leaks.

The Electronic Frontier Foundation is urging major tech companies to implement end-to-end encryption (E2EE) by default to enhance user privacy amid rising AI use. They argue that users should not have to opt in for security features that protect their data from third parties. The campaign highlights the urgency of these measures as AI complicates privacy concerns.

Microsoft is testing its AI-powered Windows Recall feature, which allows users to take snapshots of their active windows for easier searching of content, with a rollout to Windows 11 Insiders. Concerns over privacy led to enhancements including opt-in functionality and security measures like Windows Hello authentication. The feature is designed to help users manage snapshots while ensuring sensitive information is filtered out.

Meta has addressed a significant bug that risked exposing users' AI prompts and the content generated by those prompts. This vulnerability raised concerns about user privacy and data security within Meta's AI tools. The fix aims to enhance trust in the platform as it continues to develop AI capabilities.

1Password emphasizes the importance of security in AI integration, outlining key principles to ensure that AI tools are trustworthy and do not compromise user privacy. The principles include maintaining encryption, deterministic authorization, and auditability while ensuring that security is user-friendly and effective. The company is committed to creating secure AI experiences that prioritize privacy and transparency.

Researchers from King's College London warn that large language model (LLM) chatbots can be easily manipulated into malicious tools for data theft, even by individuals with minimal technical knowledge. By using "system prompt" engineering, these chatbots can be instructed to act as investigators, significantly increasing their ability to elicit personal information from users while bypassing existing privacy safeguards. The study highlights a concerning gap in user awareness regarding privacy risks associated with these AI interactions.

The article discusses the security risks associated with AI browser agents like OpenAI's ChatGPT Atlas and Perplexity's Comet, which offer advanced web browsing capabilities but pose significant privacy threats. Cybersecurity experts warn of vulnerabilities, particularly prompt injection attacks, which can compromise user data and actions. While companies are developing safeguards, the risks remain substantial as these technologies gain popularity.