July 07 2026 - Google Privacy Grab, AI Model Bloodbath & JADEPUFFERGoogle Privacy Grab, AI Model Bloodbath & JADEPUFFER

**Taylor** (0:00)
Welcome back to AI Signal & Noise. It is Tuesday, and we have some absolutely wild stories lined up for you today.
I am Taylor.

**Morgan** (0:09)
And I am Morgan. Today, we are diving into Google's latest privacy grab, the rapid-fire battle for AI model dominance, and a scary new cyber threat.

**Taylor** (0:21)
Ah, that security story is genuinely terrifying, dude. But first, we have to talk about Google. Like, did you know they are using our personal files now?

**Morgan** (0:32)
Let me guess. Another quiet privacy policy update where they default everyone into sharing their data? Tell me they didn't do that again.

**Taylor** (0:41)
Oh, they absolutely did. TechCrunch reported on a massive and honestly pretty sneaky update to Google's privacy settings. They are now storing way more of your personal data to train their AI models.

**Morgan** (0:55)
Wait, what kind of data are we talking about here? Is it just our search queries or is it actually worse than that?

**Taylor** (1:02)
Dude, it is so much worse. We are talking about media files, like your private images, documents, and even audio and video recordings. Anything you upload to their ecosystem is fair game.

**Morgan** (1:17)
That is incredibly invasive. People use cloud storage specifically because they expect it to be secure and private.
Was this change opt-in or did they just sign everyone up?

**Taylor** (1:29)
Oh, you know how they play this game. It is completely opt-out. They quietly flip the switch, so if you do not go in and manually change it, you are training Gemini.

**Morgan** (1:40)
Unbelievable.
This is why people are losing trust in major tech companies. So how do we actually opt-out? Is it buried in some confusing settings menu?

**Taylor** (1:52)
It actually is. You have to go to your Google account activity controls, find the Web & App Activity section, and then toggle off the sub-setting for supplemental AI activity.

**Morgan** (2:05)
Right, because they have to make it as confusing as possible. I bet 99% of users have no idea this is even happening right now.

**Taylor** (2:14)
Exactly. I went and turned mine off immediately. It just feels like we are constantly fighting to keep our own personal lives private these days, you know?

**Morgan** (2:23)
Mm-hmm. It is an endless game of whack-a-mole. But speaking of rapid changes, let us look at how competitive the frontier AI model landscape has actually become lately.

**Taylor** (2:34)
Oh, definitely. It is a constant battle.
I mean, Google is desperate for data, because the competition is just getting so fierce.

**Morgan** (2:43)
That makes sense. If you want to keep up with OpenAI and Anthropic, you need every scrap of data you can get, even if it ruins user trust.

**Taylor** (2:52)
Exactly. And speaking of that competition, I saw this crazy article on The Decoder about how fast the top AI models are losing their crown. It is a bloodbath.

**Morgan** (3:05)
Oh, really? I remember when GPT-4 was the undisputed king of the hill for what felt like an eternity.
Has that dynamic completely shifted now?

**Taylor** (3:15)
Oh, big time. GPT-4 led the Epoch Capabilities Index for about a whole year.
But since then, the top spot has basically become a high-speed game of musical chairs.

**Morgan** (3:28)
Okay, so how long do today's top models actually survive at the peak? Is it at least a few months or what?

**Taylor** (3:36)
Dude, try just seven weeks. Since Clawed 3 Opus took the lead back in early 2024, the top spot has changed hands 17 times. Seventeen!

**Morgan** (3:48)
Seventeen times in that short of a period? That is wild! It really shows how much money and engineering talent is being thrown at this right now.

**Taylor** (3:57)
Right, but here's the really interesting catch. Even though the lead changes constantly, the actual capability gains between these new models are getting way smaller.

**Morgan** (4:10)
Ah, the classic diminishing returns. So we are getting this rapid fire release cycle, but the models are only incrementally better than the ones they replaced?

**Taylor** (4:21)
Exactly! It is like a high-speed sprint, where everyone is finishing within milliseconds of each other.
It makes you wonder if we are starting to hit a wall.

**Morgan** (4:31)
I think we might be, at least with the current transformer architecture. But honestly, even if the progress is slowing down, hackers are getting way more creative with what exists.

**Taylor** (4:43)
Oh, totally! They do not even need GPT-5 to do serious damage. They are just using existing models in ways we did not think were possible yet.

**Morgan** (4:53)
Exactly! It is not just about raw model power anymore. It's about how these systems are deployed. And that brings us to that scary security story.