One safety story a week, pressure-tested for what's actually happening underneath the surface.

By Lindsay Gross

Everyone's talking about the AI slop flooding streaming. Spotify pulled 75 million spam tracks off the platform this year and says it isn't done, and on Deezer more than 90,000 fully AI-generated songs now get uploaded every day, more than half of all the new music going up. So for every song a real person writes and posts, a machine posts one too. Those are the numbers in the headlines. The number that actually matters is 61,026: the recordings Sony Music says it fingerprinted inside the training data of Suno, the app that builds a finished song from a one-line text prompt. Sony is moving to add them all to its copyright case in a Boston federal court in front of Chief Judge F. Dennis Saylor, where Suno has admitted its dataset held tens of millions of recordings. At the $150,000 statutory cap per song, that one motion is the difference between roughly $84 million in exposure and about $9 billion.

The flood works as a business because of how streaming pays. Spotify doesn't pay a fixed rate per stream; it splits a revenue pool, and independent estimates put a single play between a third and a half of a cent. The songwriter Tift Merritt, who co-chairs the Artist Rights Alliance, puts it plainly: a record now needs about 10 million streams just to break even. A Suno subscription runs $24 a month and turns out as many as a couple hundred commercially licensed tracks in that time. Nobody has to listen closely for that to pay. It takes volume, and a share of the same limited pool the working musician is trying to live on. That is the payment dilution Deezer's CEO, Alexis Lanternier, has been warning about for almost two years.

If you build on generative models, keep reading, because this reaches your work too. The Boston question is whether training on material you never licensed counts as fair use, and a German court in Munich rules on a version of it on July 31. The image model, code model, or text model you ship on was assembled the same way, from data you can't audit and didn't clear. Music simply has organized rights-holders and audio fingerprinting, so it reaches court first. Deezer already tags AI tracks at 99.8% accuracy and licenses that detector to other companies, so proving what is and isn't machine-made is quickly becoming standard infrastructure.

So here is the test for this week: Do you know what your model was trained on, and could you prove it if a rights-holder matched your output to their catalog the way Sony says it matched Suno's? If the honest answer is no, that exposure is already sitting in your stack. You just haven’t been sued yet.

The longer version, including where the new licensing deals between major labels and AI music companies leave working artists, is on Hidden Layer:

From idea to production: how AI-assisted development and community experimentation built a real network.

A few months ago, during the Windsurf Open Build, one of our chapter members began exploring an idea: what if professional networking could be smaller, more intentional, and actually useful?

That idea has now become Cadena Hub, an AI-powered B2B networking platform officially launching on August 1, 2026.

Cadena Hub is designed around curated, high-signal connections. Members are placed into small groups (“hubs”) that refresh monthly, and each week they receive a deliberate one-to-one introduction: complete with a written rationale for why the match was made. Contacts unlock through these introductions, creating a network that grows through context rather than cold outreach.

The platform supports a full member experience: onboarding, hubs, introductions, messaging, meetings with built-in video rooms, and a collaboration space. Behind the scenes, an admin portal handles application review and member management.

What makes this especially interesting is how it was built.

The project started as an early prototype during Windsurf, then evolved through a combination of Windsurf tooling and Claude as the primary development agent. When the project changed hands earlier this year, the new technical direction focused on something critical: verifying reality at every step.

Instead of relying on what an AI system “believed” it had done, the build process enforced strict validation:

This approach was paired with a tool-driven workflow:

The result is a system where AI accelerates development, but never replaces verification.

By July 21, the platform had already processed 140 applications, with over 130 approved members and 137 companies onboarded ahead of launch. The first cohort goes live August 1, when the initial hub rotation runs.

Cadena Hub is a strong example of what can emerge from combining community-driven experimentation with modern AI tooling: an idea formed in an open build environment, iterated with AI agents, and matured into a production-ready platform with real users.

We're thrilled to share that The AI Collective is a proud community partner of the AI Summit at Stanford, and honestly, this is the event to be at this summer.

Happening July 30 to August 1, this pioneering residential AI summit brings together the leading founders, investors, technologists, and researchers shaping the future of AI, all on Stanford's campus, for three unforgettable days.

The speaker lineup currently includes:

And that's just the start. This won’t be a hotel-ballroom conference; it's three days of living on Stanford’s campus alongside the people building and funding what comes next.

Our partners and sponsors get exclusive placements across the newsletter and access to AIC's in-person network — demo nights, dinners, hackathons, and forums across 180+ chapters.

For all inquiries, send us a note at [email protected].

The AI Collective is built by volunteers across 180+ chapters in 40 countries.

Thank you to the thousands of volunteers around the world who make this work possible. We truly could not do this without you.