Google Home lets outside AI agents into the house, cameras and all

Google has begun rolling out a Model Context Protocol integration that lets third-party AI agents work with Google Home devices and event history. The practical shift is bigger than another chatbot gaining a thermostat button: compatible agents can summarise camera activity, inspect household patterns, trigger automations and build custom dashboards, although early access is currently limited to Google Home Premium Advanced subscribers in the United States.

Watch Desk analysis

What happened

The Ambient reports that Claude, OpenClaw, Hermes and Google Antigravity are among the compatible systems. Google Home MCP can reach connected devices and stored activity, with examples including thermostat control, camera summaries, custom automations and personalised home-control interfaces.

The feature is being offered as an early-access rollout, so bugs and missing functionality are part of the bargain. Google is asking developers to report problems through its smart-home feedback channel. The source also notes that third-party smart-home AI is not entirely new, with Home Assistant already supporting similar connections through community guides.

Why it matters

This turns the smart home from a walled garden into something closer to a shared operating environment for agents. That could make useful tasks easier, such as asking one system to summarise what happened across several cameras or assemble a dashboard without dragging icons around for half an hour.

It also means an agent may be given access to far more than a light switch. Camera feeds, device history and household routines are intimate data, and prompt-injection attacks could hide malicious instructions in ordinary emails or calendar entries. Convenience has arrived carrying a clipboard labelled “permissions”.

Our read

Google is making a meaningful move by opening Home to agents beyond its own assistant, but early access is not the same as a settled safety model. Readers should treat the integration as an experiment, check which devices and histories an agent can reach, and avoid granting broad access merely because a setup screen makes it look tidy.

The useful test is not whether an agent can turn off the lights. It is whether people can understand, limit and audit what happens when the agent is wrong, manipulated or simply overconfident.

What to watch

• Access controls:
  whether Google adds finer permissions for cameras, history and device actions.
• Prompt injection:
  whether researchers demonstrate attacks against connected-home agents and Google patches them.
• Availability:
  whether the feature expands beyond US Premium Advanced subscribers.
• Independent testing:
  whether outside developers find the safeguards robust in real homes. Source: The Ambient, which reports the rollout and its early-access limits.

Discussion spark: Should smart-home platforms expose cameras and household history to any compatible AI agent by default, or should every sensitive device require separate permission?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.