Op please change the title as to not further fuel the misinformation. Arstechnica seems to enjoy burying the actual information 3 paragraphs in where they know nobody will read them and your embellishments aren't helping.

1. There's nothing anywhere that suggests the ai is "remote controlled".

2. Arstechnica suggests that humans have access to the data and as evidence they linked a site saying humans don't have access.

3. Ars seems to want us to think the red rectangles in the image are contradictory. They aren't as Gemini Apps can be individually turned off independent of Gemini Apps Activity (history) and vice versa. The forced 3 day activity storage doesn't enable the apps themselves.

4. Even the author of the article pointed out that it can be turned off

5. Ars cites Tuta's article selectively to make it sound like disabling gemini is either ineffective or complicated. Tuta itself is a privacy focused gmail alternative with vested interest to muddy the waters but their article is still somewhat better written than Ars'

Edit: Title when this comment was written for posterity:

there is a remote controlled ai agent on every google device that can not be turned off

Pretty sure both have been available through portals for at least a year

Im not sure I like the cpu utilization being based on the base clock. A percentage that can go arbitrarily above 100 doesn't sound useful for determining bottlenecks. It must be difficult to accurately determine given all the dynamic boost stuff but since all other such utilities figured it out surely they can too. Hope they don't just leave this as good enough.