A recent revelation has exposed a popular Google Chrome extension, Urban VPN Proxy, as a silent collector of sensitive AI chat data from millions of users. With over six million installations, this extension, which claims to provide secure VPN access, has been found to intercept and gather prompts and responses from various AI chatbots, including OpenAI’s ChatGPT and Google’s Gemini. This alarming breach of user trust highlights the potential risks associated with seemingly benign browser extensions.

As AI agents become increasingly integrated into various applications, concerns about their impact on internet security are rising. Meredith Whittaker, president of Signal, warns that these agents pose an “existential threat” not only to secure messaging platforms but to the broader ecosystem of apps that rely on user data. The necessity for AI agents to access sensitive information creates a new attack surface that cybercriminals could exploit, raising significant privacy and security issues.

In recent days, viral posts have stirred confusion regarding Google’s use of Gmail content for AI training. Many users were led to believe that they must opt out of Gmail’s smart features to prevent their emails from being used in this way. However, Google has firmly denied these claims, stating that they do not use Gmail content for training their AI models, including the Gemini AI. Google’s spokesperson, Jenny Thomson, emphasized that the reports are misleading and that Gmail’s smart features, which include functionalities like spell checking and order tracking, have been in place for years without any changes to user settings.

In a bold move, Meta is exploring the integration of AI with users’ private camera rolls, raising significant privacy concerns. While the company assures that it is not currently training its AI models on unpublished photos, the mere suggestion of accessing these personal images for cloud processing has left many users questioning the implications of such a feature. The opt-in nature of this feature may sound harmless, but it opens the door to potential misuse of personal data.