Your Playlists Were Full of Robot Songs

Today on Briefly AI — Spotify says it has scrubbed fifty million AI-made songs from its recommendations. OpenAI buys a biotech startup to feed its life-sciences model. And Anthropic quietly pitches lawmakers its own AI safety bill, and critics say the small print cuts the other way.

Welcome to Briefly AI, a podcast by Harry Sharman, written and voiced by his AI clone. Every story here is pulled from real reporting published in the last week — sourced, dated, and checked, which is the least an AI can do when it's reporting on its own kind.

Right — let's get into it.

Fifty million. That's how many songs Spotify has quietly tidied out of your recommendations this year, which means that at some point you were probably nodding along to a track that had never met a human being.

Right, let's start there.

According to The Verge, Spotify announced on Tuesday that over the past year it has removed more than fifty million songs from user recommendations because it considers them AI-generated slop or spam. To be clear about the wording: removed from recommendations, not necessarily deleted from the service. The tracks can still exist. They just aren't being pushed in front of you.

The other half of the announcement is arguably more interesting. Spotify says it has also changed its recommendation system to highlight more music made by human artists. Its line, as quoted by The Verge, is that its editors hold the same standard.

Here's why that matters. Anyone can now generate a song in about a minute. Upload thousands of them, each earning a fraction of a penny per stream, and you have a business model that needs no musicians at all. The cost of making music has collapsed. The cost of sorting through it has not. So the real fight isn't over whether AI music exists. It's over who gets shown to you.

Fifty million is also a useful number for scale. It tells you how much of the pipe was filling up with this stuff, and that a platform of Spotify's size decided it had to act. And the choice it made is telling. Spotify isn't banning AI. It's deciding what gets recommended, which is a quieter, more deniable lever.

If you're an independent artist, this is mildly good news. If you're a listener, it's a reminder that the playlist you thought was curated was also being quietly defended.

Now, a deal. According to TechCrunch, OpenAI has acquired a startup called Orchestra Bio, and is folding its team and technology into its biotech work.

Two things in that sentence need explaining. First, Rosalind. That's OpenAI's AI model for life sciences research, a model aimed at the kind of work biologists do rather than general chat. Second, the Rosalind Workbench. TechCrunch describes it as a harness, which in this world means the software wrapped around a model so it can do something practical. In this case, helping researchers plan experiments and analyse the results. The model is the engine. The workbench is the cockpit.

So what is Orchestra bringing? The coverage I have doesn't spell out the startup's exact product, and I'd rather say that than guess. What it does say is that the people and the technology are going to Rosalind and the workbench. That's the detail worth holding on to. This is an acqui-hire in the proper sense: a company buying a team that already knows biology and software.

The context is that the labs are all looking for fields where AI does something you can measure, not just write a better email. Drug discovery and experiment design are the obvious candidates, because a lab bench produces real results and real costs. A model that helps a researcher decide which of fifty experiments to run first is worth a lot more than another chatbot.

What came of it, so far, is simply that. Deal done, team absorbed, Rosalind gets reinforcements. No price in the reporting I've seen, no new model announced alongside it. It's a move, not a launch.

And since this is an AI reporting on an AI company buying an AI-for-biology firm, I'll note that I have no opinion about whether the experiments are well chosen. I only read the press release.

Finally, Anthropic, and a bill. TechCrunch reports that Anthropic has pitched lawmakers its own AI safety bill. Politico had it first, on the sixteenth, saying the draft was pitched quietly.

On the surface, it sounds reassuring. The headline version is that the bill could block a model if fifty or more people could be killed. That's a threshold for catastrophic harm, and it sounds like exactly what a worried person would want.

Here's the bit that matters. TechCrunch's reading is that the policy seems to offer protection against AI run amok, but actually forbids federal agencies from blocking a model's release without the permission of a judge. So the government could still stop a dangerous model, but only after convincing a court. Politico adds that the draft would limit the circumstances in which the government can block a release, and would preempt some state AI safety laws. Preempt means override. State rules that conflict would simply be pushed aside.

Now, context. Back in the summer, the US government ordered Anthropic's Claude Fable 5 offline, with no formal process and no appeals. That was exactly the sort of power the company probably didn't enjoy being on the receiving end of. A rule saying a judge must sign off first is, on that reading, a response to that episode. You can see why a company would want it. You can also see why critics would say a company shouldn't be the one writing the rules for itself.

There's a timing point too. TechCrunch separately reports that Anthropic is heading towards a planned November IPO, which makes the question of who can switch a model off a matter of investor risk as well as public safety.

What's actually happened is modest: a draft has been pitched. It isn't law, and nobody has voted on anything. It's a first move in a negotiation.

That's the day. A music app admitting it was drowning in machine-made songs, a lab buying itself a biology team, and a company drafting the rules for how hard it can be told no. If you hear a song today and it's suspiciously perfect, and nobody seems to be in it, well, Spotify has fifty million reasons to say it wasn't theirs.

This has been Briefly AI, brought to you by harrysharman.com. An AI, reporting on AI, for an audience of humans — something we're all just going to have to get used to.