June 12 2026 - Deezer AI Scans, Perplexity & Anthropic FalloutDeezer AI Scans, Perplexity & Anthropic Fallout

**Taylor** (0:00)
Happy Friday, everyone! Welcome to AI Signal & Noise, your weekly wrap-up of the wildest stuff in artificial intelligence. I'm Taylor, and I am so hyped for today's show.

**Morgan** (0:14)
And I'm Morgan. Don't get too carried away, Taylor. We've got some pretty serious and honestly slightly concerning developments to unpack today.
It's not all fun and games.

**Taylor** (0:24)
Oh, come on, Morgan, it's Friday!
But yeah, okay, you're right. We have some massive updates from AI music detectors to some super shady behavior from Anthropic.

**Morgan** (0:36)
Exactly. Plus some fascinating agent safety research from DeepMind.
Let's start with the music, because this one is actually pretty wild.

**Taylor** (0:45)
Dude, yes, so Deezer just dropped this free tool that lets you scan your playlists on Spotify, Apple Music or wherever to see if they have AI generated tracks hiding in them.

**Morgan** (0:57)
Wait, so it's not just for Deezer users? Anyone on any major streaming service can use this to audit their own playlists? That's actually a really smart move.

**Taylor** (1:06)
Right? It's like totally open to everyone. I tried it and it's so easy.
Apparently, they want to give listeners more transparency about what's actually human made.

**Morgan** (1:18)
But wait, how accurate is it really? AI detection in text is notoriously unreliable. So, is music detection actually any better? Or is this just a PR stunt?

**Taylor** (1:29)
Well, Deezer says their tech is super advanced at spotting the specific digital signatures of AI generators. They want to protect artists' rights, you know?

**Morgan** (1:39)
I get that, but what happens if it flags a reel in the artist who just used some digital synth or a weird production tool? False positives could ruin someone's career.

**Taylor** (1:50)
Oh man, I didn't even think about that. That would be awful.
But like, on the flip side, isn't it good to stop spammy AI tracks from taking royalty money?

**Morgan** (2:02)
Oh, absolutely. The streaming platforms are drowning in AI generated white noise and generic lo-fi beats.
If this helps clean up the library, I'm all for it.

**Taylor** (2:13)
Exactly. It's about giving us a choice. I want to know if I'm vibing to a human or a server rack, you know?
It's a crazy time for music.

**Morgan** (2:23)
It really is. But speaking of massive computing power, did you see what Perplexity is doing with their new deep research update? It's kind of mind-blowing.

**Taylor** (2:33)
Oh, dude, yes. Perplexity is moving deep research directly into something they call Perplexity Computer. It basically breaks hard questions down into subtasks.

**Morgan** (2:45)
Right. And instead of just using one model, it routes those subtasks across more than 20 different frontier models. That sounds incredibly complex and expensive.

**Taylor** (2:56)
It's insane. It can generate entire reports, slide decks, and even dashboards. It's like having a whole research department running in the background while you grab coffee.

**Morgan** (3:09)
But is it actually useful or is it just throwing brute force compute at a problem? I mean, 20 models. How do they even coordinate all of those outputs?

**Taylor** (3:18)
Apparently, the system acts like a manager. It assigns the best model for each specific task, like math to one, writing to another, and then compiles it all.

**Morgan** (3:31)
Okay. That part is actually brilliant, if it works. It solves the jack of all trades, master of none issue we see with single models. But what about the latency?

**Taylor** (3:42)
Well, yeah, it's not instant, but for a deep report or a slide deck, waiting a few minutes is nothing compared to spending hours doing it yourself, right?

**Morgan** (3:53)
Fair point. I just worry about the hallucination rate. If 20 different models are contributing, that's 20 potential points of failure for incorrect data.

**Taylor** (4:03)
Totally. But they claim their routing system cross-checks facts. I guess we'll have to see if the decks it makes are actually client ready, or just pretty garbage.

**Morgan** (4:13)
Yeah, we'll definitely be testing that. But all this talk about multiple models working together brings us to some pretty scary research from Google DeepMind.

**Taylor** (4:22)
Oh man, this one gave me total sci-fi vibes. DeepMind is literally funding research because they are terrified of what happens when millions of AI agents start talking to each other.

**Morgan** (4:36)
It's a very real concern. Rowan Shaw from DeepMind pointed out we are heading toward a mass market arrival of autonomous agents that don't have human oversight.

**Taylor** (4:46)
Like, they'll just be taking instructions from other AI agents. It's like a giant game of digital telephone, but with real world consequences, dude.

**Morgan** (4:57)
Exactly. Imagine an agent booking a flight, talking to an airlines agent, which talks to a pricing agent.