In 2024, Spotify took enforcement action against 87 accounts for the kind of abuse we are about to discuss. In 2025, that figure was roughly 3,500. By the time a US Senate inquiry forced the company to fully account for itself in June 2026, Spotify had removed more than 57,000 fake podcast episodes — spread across over 3,000 shows — that used AI-generated audio to funnel listeners toward illegal online pharmacies and cryptocurrency scams.
That is a jump of more than 40x in actioned accounts in a single year, and a five-figure pile of synthetic episodes built on top of it. The drugs on offer were not abstract: modafinil, opioids, stimulants, benzodiazepines, sold through unregulated storefronts with names like opioidstores.com — some of which the DEA later seized on its own. The episodes themselves were often just a few seconds of robotic narration, a throwaway shell whose only real payload was a link, a phone number, or a search-engine-friendly title stuffed with the name of a controlled substance.
The instinct is to read this as a drug-enforcement story. It is more useful to read it as a story about what happens when generative AI is pointed at a platform’s trust infrastructure — and how little stood in the way.
The economics of synthetic spam
What changed between 87 and 3,500 was not the criminals’ ambition. It was their unit cost. Producing a fake podcast used to require, at minimum, a human reading a script. Generative audio tools collapsed that cost to nearly zero. Once an episode costs nothing to make, the optimal strategy is to make an enormous number of them and let volume do the work — flood the catalog, capture a sliver of search traffic, and route whoever clicks toward an external storefront.
Spotify’s own data tells you the spammers understood this perfectly. The company reported that 94% of the removed episodes received zero plays, and 99% drew fewer than ten streams. The operators were not trying to build an audience. They were treating Spotify as a free, high-authority web host — a place to park keyword-stuffed pages that rank well in Google precisely because they sit on a trusted domain. The podcast was never the product. Spotify’s credibility was the product, borrowed and resold.
This is the part worth sitting with. The abuse did not exploit a bug in Spotify’s code. It exploited the thing that makes Spotify valuable: that a listing on its platform carries an implicit endorsement, that its domain confers legitimacy, that its search results are assumed to point somewhere real. AI spam is corrosive not because each episode is convincing, but because the sheer quantity overwhelms the assumption of curation that a platform’s reputation rests on.
A moderation gap by design
Spotify’s response to Senator Maggie Hassan’s office included a striking admission: the company is “not particularly well-positioned” to identify AI-created podcast content. It runs automated moderation for music, but no equivalent system exists for podcasts. And because it does not track whether users click the hyperlinks embedded in episodes, it has almost no visibility into how many people the spam actually reached on the way to a pharmacy checkout page.
That gap is not an oversight so much as a consequence of how these systems were built. Recommendation and search engines are optimized to surface content and lower the friction of discovery. They are not, by default, optimized to ask whether the content should exist at all. For years that asymmetry was tolerable, because producing bad content at scale was expensive. Generative AI removed that natural rate limit, and the moderation side of the ledger had no equivalent acceleration. The 40x surge in enforcement is what it looks like when a problem that used to be self-limiting suddenly is not.
It is also telling that none of the 57,000 removals triggered a referral to law enforcement, even when episodes linked to sites peddling opioids. Spotify framed the episodes as a “spam attack” rather than an attempt to sell drugs to its users — a characterization that is technically defensible and strategically convenient. If it is spam, it is a content-quality nuisance. If it is a drug-distribution channel riding on your infrastructure, it is something you may have an obligation to report. The language matters because it determines who is responsible.
The privacy and safety cost
There is a direct safety harm here — real people steered toward unregulated opioids and benzodiazepines by audio that mimics the cadence of a legitimate show. But the deeper damage is to the information environment itself, and that is where this becomes a privacy and trust story rather than only a drug story.
Every recommendation system is a promise about relevance: that what you are shown bears some honest relationship to what you searched for. AI spam at this scale breaks that promise quietly. You do not see 57,000 fake episodes; you see one that happened to surface, and you have no way to know whether it is the product of a creator or a content farm. The same dynamic that let scammers launder credibility through Spotify is now playing out across every platform that hosts user-generated media — app stores, video sites, review systems, search results. The defining security problem of the generative era is not a clever exploit. It is volume: the cost of producing plausible content has fallen below the cost of vetting it.
Spotify deserves some credit for eventually moving at scale, and for disclosing the numbers when pressed. But the timeline is the indictment. The 3,500 bans came only after CNN documented the pipeline in May 2025, and the full 57,000-episode purge came only after a senator demanded answers. Enforcement was reactive, triggered by journalism and political pressure rather than by the platform’s own detection. A system that can only find the abuse after a reporter points to it is not really a defense. It is a cleanup crew.
The number to remember is not 57,000. It is the gap between 87 and 3,500 — the speed at which an abuse vector can go from marginal to overwhelming once AI removes the cost of producing it. That curve is not unique to Spotify, and there is no reason to think it has finished climbing.
Sources: The Next Web, CNN Business, Inside Radio, Senator Hassan / JEC.
