Guest column: Music Ally publishes guest columns which voice the opinions of notable authors that advance specific perspectives on important issues. These are chosen at the Editorial team’s discretion and are not paid-for. You can explore the guest column archive here.
This is a guest post by veteran tech and music executive Virginie Berger.
DeepSeek dropped its V3 model, DeepSeek-R1 and DeepSeek-R1-Zero on 26 December, alongside its paper explaining their breakthrough technology, and I spent my Christmas break playing with it. Yes, really.
Then, on 20 January, DeepSeek launched its DeepSeek-R1-Distill models: a low-cost, local and open-source Large Language Model (LLM). Since then, two other models have been launched, KIMI 1.5 and Qwen 2.5 VL, all agentic multimodal models from China.
China’s top AI model isn’t even DeepSeek; it has at least ten top-tier models trained from scratch, surpassing Europe’s best (sorry, Mistral). The US, meanwhile, has just five major players: OpenAI, Anthropic, Google, Meta, and xAI.
I’ve been using DeepSeek for over a month now, and it’s both awe-inspiring and mildly terrifying. I even brought it up during my AI licensing panel at the recent Music Ally Connect Conference because, let’s be honest, it’s a looming existential threat to the music industry.
This is precisely what I warned about in my December Forbes article on AI’s impending storm for 2025 and again in my latest piece on AI geopolitics and copyright erosion. Spoiler: this is it, the storm has arrived.
DeepSeek R1 spells out an inevitable future: AI that is cheap, commoditised, and ubiquitous. Machines training machines. The automation of jobs at a scale and speed we are woefully unprepared for.
AI is a looming existential threat to the music industry
The DeepSeek R1 algorithm works. And it’s spreading. You can download the code, run it on your own server or PC, and see results that differ from those on Chinese-hosted versions. DeepSeek means that everyone, from researchers in São Paulo to start-ups in Stockholm and doctors in Nairobi, can access state-of-the-art AI at little to no cost.
It means that someone in a remote hospital can train a local AI model and use it to create its own LLM to find patterns in patient data, for instance. And being local means they don’t have to rely on sharing their data externally.
You just need a $2,000 machine with 512GB RAM to run DeepSeek R1 locally, generating 3.5–4 tokens per second. Fully offline, fully private. I could vanish into the woods and still carry the sum of human knowledge in a box. But, alas, unresolved copyright issues remain the elephant in the room.
Cheap, commoditised and ubiquitous
For the music industry, DeepSeek’s simple architecture accelerates AI system development. Machines training machines at this scale mean an imminent flood of generative AI tools, many relying on unethically-sourced data.
DeepSeek itself was trained on mass copyright violations: on its paper for its Vision-Language (VL) model there is a list of the training data used, including a surprising mention of Anna’s Archive, a collection of pirated copyrighted works. “We cleaned 860K English and 180K Chinese e-books from Anna’s Archive,” explains the paper. Open-source AI isn’t automatically ethical, and the label shouldn’t distract from the questionable training methods behind it.
DeepSeek makes AI-generated music even more accessible, leading to increased competition for human artists, algorithmic manipulation, and shrinking royalties. AI-generated tracks can be optimised to dominate streaming recommendations, diverting revenue from real musicians.
Imagine developers training a local AI model on 20 years of pop music and flooding Spotify with 10,000 AI-generated tracks daily. Open-source AI also removes accountability: anyone can modify music models to bypass copyright detection, making enforcement nearly impossible.
AI quantum computing and AI agents will further disrupt the industry. AI-powered agents can create new music so quickly and in such massive quantities that copyright enforcement becomes meaningless.
AI quantum computing can reproduce copyrighted styles so accurately that infringement claims become futile and can train on protected works without detection, further accelerating music plagiarism. Autonomous AI agents can function as record executives, controlling which artists get promoted, signing contracts, and even dictating music trends, all without human oversight.
At this scale, copyright enforcement may become meaningless, and the entire role of intermediaries, like digital distributors, will be wiped out.
Open-source AI also removes accountability: anyone can modify music models to bypass copyright detection
Meanwhile, the UK is busy gutting copyright protections to chase some fantasy of being an “AI superpower.” DeepSeek crushed that notion in under two weeks. Why build billion-dollar data centers when a few open-source models deliver the same results?
I’m now channeling my inner Alanis Morissette. According to OpenAI CEO Sam Altman and the new US AI Czar David Sacks, DeepSeek may have stolen American AI IP through “distillation,” a technique where a student model extracts knowledge from a parent model.
Isn’t it ironic, don’t you think, considering OpenAI built its empire by scraping the internet without permission, that it is now whining about DeepSeek distilling its outputs.
The real kicker? By framing its battle with DeepSeek as an IP violation, OpenAI is effectively asserting ownership over AI-generated content. Let that horror sink in. If AI-generated text, images, music, and code become default, who owns what? And what happens when AI overlords decide you’re using “their” outputs the wrong way?
There’s even a new bill in the US congress called the “Decoupling America’s Artificial Intelligence Capabilities from China Act of 2025” that would ban the import of any AI technology from China, including open-source models like DeepSeek.
The bill would also make it illegal to do any research or development in AI in collaboration with an “entity of concern”, defined as any Chinese institution or company. Oh, and the “content” belongs to the system. We’re on the brink of techno-feudalism, where a handful of AI giants dictate who gets to innovate.
Irony check: here’s what the AI grifters were saying last week: “If AI learning is theft, then all of humanity are thieves. Anti-AI sentiment is Luddite elitism. AI democratises talent. Adapt or die.” And the AI grifters this week? “China stole our AI!”.
You can’t cry about China “stealing” your data when it was never yours to begin with. You can’t claim China stole research you never published. You can’t cry about censorship when your own models are just as filtered.
A wake-up call
DeepSeek isn’t just another AI model, it’s a wake-up call. The music industry is sitting on a goldmine of data, yet we’re allowing AI to exploit it unchecked. The real power isn’t in the models themselves, but in the data and metadata that fuel them.
Yes, some stakeholders are taking action. Majors are suing Suno and Udio in the US; GEMA is suing OpenAI in Germany; and there are local campaigns urging policymakers to intervene. But these efforts are limited, reactive, and fail to address the deeper problem: the industry itself is divided and, in many cases, actively embracing the AI narrative instead of confronting the structural issues.
We’re coddling Big Tech narratives about curing cancer and creating utopias while they systematically steal from creators. Machines are training on stolen work, and if we don’t fight back now, there won’t be anything left to protect.
Plus, the threat is escalating. AI agents are already here, and in less than two years, AI quantum computing will upend everything we know. The music industry can’t afford to waste more time in endless debates while AI companies push forward unchecked.
Key players in the music industry remain hesitant, whether out of fear, complacency, or opportunism, to take a unified, aggressive stance against AI systems built on copyright theft.
It’s time to shift the conversation from passive negotiations to active resistance. The industry must invest in defensive technologies, tools that block AI training, disrupt unauthorised scraping, and create AI “tar pits” that trap exploitative models before they can do further damage.
The industry also needs a clear, collective and global push for real AI regulations and robust enforcement mechanisms. Otherwise, it’s just empty words while the foundations of creative ownership are being dismantled in real time.
The choice is clear: either we take control now, or we watch as AI reshapes the industry without us.
