Google Brings AI Agents Into Smart Homes, Raising Privacy Risks

Google has announced formal Model Context Protocol (MCP) support for Google Home, allowing compatible AI agents—including Google Antigravity, Claude, Hermes and Open Claw—to access connected devices and event-history records. Google says the integration will enable natural-language tasks such as summarising camera activity across rooms, identifying trends in home activity, controlling appliances and creating customised monitoring dashboards from existing device data.

The feature is expected to enter early access in the coming weeks. Initially, it will be restricted to Google Home Premium Advanced subscribers, which costs $20 (£15.60) per month or $200 (£156) per year, and setup requires a Google Cloud project with MCP permissions. Google says it has retained safeguards, including blocking sensitive actions such as unlocking doors, while access can be revoked through the Google Home app.

The move also raises privacy and security concerns because AI agents could receive highly sensitive household data and control smart-home systems. The article cites concerns about MCP’s lack of mandatory authentication and refers to previous data-leak incidents involving the protocol, although it provides no specific incident details.

Google’s documented controls are therefore the main practical protections identified: users considering early access will need to create the required cloud project, grant permissions and use the Google Home app to withdraw access when necessary.