Google is expanding its push into desktop AI, reportedly testing a dedicated Gemini app for macOS with a new feature called ‘Desktop Intelligence’ – as separate security revelations highlight how quickly personal data can be exposed on consumer devices.

According to reports, Google has begun limited external testing of a native Gemini app for Mac, bringing it into direct competition with OpenAI’s ChatGPT and Anthropic’s Claude, both of which already offer desktop clients.

While Gemini is currently browser-based for most users, a standalone app would streamline access and deepen integration with local workflows.

The headline feature, Desktop Intelligence, signals a broader shift in how AI assistants operate. Instead of relying solely on user prompts, Gemini would be able to interpret on-screen activity and pull context from apps such as calendars, documents and browsers. The aim is a more proactive assistant – one that understands what users are doing in real time and responds accordingly.

This aligns with a wider industry trend toward “personal intelligence,” where AI systems tap into emails, files and other personal data to deliver more relevant outputs. Google has already begun integrating Gemini across Workspace and its consumer services, while rivals are pursuing similar cross-app capabilities.

However, the growing reliance on deeply integrated AI comes as security researchers warn of a major iPhone vulnerability affecting older software.

A newly uncovered hacking technique known as ‘DarkSword’ can reportedly extract sensitive data, including messages, passwords, photos and even crypto wallet credentials, simply by luring users to a compromised website.

The attack is “fileless,” meaning it hijacks legitimate system processes and leaves little trace after execution.

The exploit targets devices running certain versions of iOS 18, with estimates suggesting a significant portion of users may still be exposed. While Apple has since patched the underlying vulnerabilities in newer updates, researchers warn that the tool’s leaked source code could allow broader misuse.

The juxtaposition is stark. As tech giants race to embed AI deeper into everyday computing, the attack surface for personal data continues to expand.