What Is Poisonify? Data-Poisoning Music vs AI (2026)

Poisonify is an adversarial audio tool that lets musicians data-poison their own tracks so AI models choke on them during training. It was built by producer and YouTuber Benn Jordan — not a startup, not a SaaS, a working musician who got tired of watching his catalog get scraped. By the end of this you'll know exactly what it does, whether it actually works, and whether it's worth your time versus just building a real business.

What Is Poisonify

Poisonify is a technique for adding inaudible adversarial noise to a music file so that AI models trained on it produce broken, unusable output — while human listeners hear no difference. It was created by Benn Jordan (who releases music as The Flashbulb) and demonstrated in his video The Art of Poison-Pilling Music Files. It draws on the same family of ideas as academic projects like HarmonyCloak and image tools like Glaze and Nightshade.

The short version: it doesn't stop anyone from listening to your song. It tries to make your song toxic to any model that tries to learn from it.

How Poisonify Actually Works

Here's what's happening under the hood, in plain producer terms.

What adversarial noise is

AI music models don't "hear" music the way you do. They learn statistical shortcuts — specific spectral features they associate with a genre, an instrument, a vocal texture. Adversarial noise targets those shortcuts directly. You add a layer of perturbation that sits below the human hearing threshold but lands right on the features the model relies on. To your ears: nothing changed. To the model: the data is garbage, and it learns garbage.

Benn Jordan reportedly demonstrated poisoned tracks making AI services like Suno hang or crash when trying to process them. That's the goal state — not subtle degradation, full refusal.

The "analog loophole" advantage

This is the part I find genuinely clever. Because the poison lives in the audio itself, it survives anywhere sound goes. Re-encoding, streaming compression, even playback and re-recording — the perturbation rides along. One writer covering Jordan's work noted this family of techniques works on audio directly, so it covers the analog loophole: it functions anywhere the sound is heard. Most protection schemes break the moment a file gets transcoded. This one is harder to strip.

The cost: compute, and a lot of it

Here's the honest catch. Poisoning audio is computationally brutal. Reporting on Jordan's work cites roughly 400 minutes of compute for every 1 minute of audio to process a track. That's not a typo. A three-minute song is a serious processing job. This is the single biggest reason Poisonify reads more like cutting-edge research and proof-of-concept than a click-and-go product you'll be running on your whole catalog tonight.

Poisonify vs Poison Pill vs HarmonyCloak — don't confuse them

This space is small and the names blur together, so let me separate them clearly because it matters:

• Poisonify — Benn Jordan's project. An algorithm/technique, demonstrated publicly, rooted in his own adversarial-noise research. Jordan was listed as a speaker at the 2026 NAMM Show, where the project is described as an algorithm that protects music from AI training models.
• Poison Pill — a separate UK startup from Ben Bowler (previously Chew·tv, Aux), launched in beta in late 2025. Same goal, different people, different product. Founder Ben Bowler has stated the aim is to protect about 20% of independent music — enough to shift the power dynamic and force AI companies toward fair licensing instead of silent scraping. Update: as of April 2026, Ben Bowler announced Poison Pill is shutting down.
• HarmonyCloak — the academic foundation, out of the University of Tennessee Knoxville and Lehigh, presented at the IEEE Symposium on Security and Privacy. Most commercial efforts build on this research.

If you see "$3/month for up to 10 songs" quoted around this topic — that's pricing associated with the startup side of this space, not confirmed pricing for Poisonify itself. I'm flagging that because half the blog posts out there merge these three into one thing, and they're not.

Quick Comparison

Best overall for most producers right now: none of these are a finished consumer product yet — Poisonify is the most interesting technically, but if you want something usable today, the startup route (Poison Pill) is closer to plug-and-play. If you only try one thing, start by understanding the tradeoff before you poison a single file.

Tool	Who's behind it	What it is	Maturity	Best for
Poisonify	Benn Jordan	Adversarial-noise algorithm/demo	Research / proof-of-concept	Technical producers who want to understand the method
Poison Pill	Ben Bowler (UK startup)	Productized poisoning service	Shut down (Apr 2026)	—
HarmonyCloak	UT Knoxville / Lehigh	Academic "unlearnable music" system	Research paper	Understanding the underlying science
Glaze / Nightshade	Univ. of Chicago	Image poisoning (not audio)	Released	Visual artists / reference point

3 Questions That Tell You If Poisoning Is Worth It For You

1. Is your music actually being scraped at a scale that hurts you? If you have 200 monthly listeners, the realistic threat to your income isn't a model training on your three loops. It's obscurity. Be honest about which problem is bigger.

2. Can you absorb the compute cost and workflow friction? At hundreds of minutes of processing per minute of audio, poisoning your full catalog is a real commitment. If that time competes with finishing music or building offers, the math may not favor poisoning.

3. Are you okay with an arms race you might not win? AI companies can theoretically train models to detect and ignore poison. The defenders can update methods in response. It's an ongoing fight, not a one-time fix. If that uncertainty stresses you out, this isn't your tool yet.

And if you're still figuring out how AI fits into your production workflow, that context is worth having before you decide whether protecting your catalog from it is a priority.

Frequently Asked Questions

What is Poisonify and who made it? Poisonify is an adversarial audio tool that adds inaudible noise to music so AI models trained on it produce broken output. It was created by musician and YouTuber Benn Jordan, also known as The Flashbulb, who demonstrated it in a video on poison-pilling music files.

Does data-poisoning affect how my music sounds? The intent is no audible difference for human listeners — the perturbation sits below the hearing threshold. The disruption only shows up when an AI model tries to learn from the file. In practice, always A/B your poisoned master against the original before release.

Is Poisonify free, and how much does it cost? Poisonify is presented as Benn Jordan's research and demonstration rather than a launched paid product, so there's no confirmed consumer price. Pricing figures like "$3/month for 10 songs" circulating online are associated with separate startups in this space, not Poisonify specifically.

Can AI companies just bypass poisoned music? Potentially, yes. This is the core arms-race problem: companies could train models to detect and remove the poison, while creators update their methods in response. Audio poisoning covers the analog loophole, which makes it harder to strip, but no method is permanent.

Should I poison my music or focus on my business? For most independent producers, business systems, fan relationships, and strong offers protect your income more reliably than poisoning your catalog. Poisoning is a defensive layer, not a revenue strategy. Treat it as optional insurance, not a priority. If you're releasing AI-assisted music, understanding how AI music distribution works is probably a higher-leverage use of your time.

The Bottom Line

Poisonify is a genuinely smart piece of adversarial research from Benn Jordan, and the bigger movement around it — Poison Pill, HarmonyCloak — is worth watching closely. But as of 2026 it's closer to a powerful proof-of-concept than a finished tool you should be running on everything you make. If you're a technical producer who wants to understand and experiment with the method, dive in. If you're trying to build a career, poisoning your files won't pay your rent — your offers and your audience will. Protect the work, but build the business first. At Beatonomy we'll keep tracking this as it matures. More on building a defensible independent career in the Autonomy section, and if you haven't yet, read think like a media company — that mindset protects you more than any algorithm.