Artists face steep income decline due to AI, UNESCO finds — what that means for the AI industry

UNESCO’s new global monitoring report says generative AI could shave nearly a quarter off some creators’ incomes by 2028, and the shockwaves run straight into the business models of AI developers and platforms.

A street-level scene helps explain the stakes: a mid‑career session musician checks a royalty statement and finds fewer dollars than last year while an AI model quietly repackages the same chord progression into a playlist that loops in airports. The mainstream read is familiar and urgent — artists are losing money to automation. The sharper business angle is that those revenue losses point to a fast growing, underpriced externality that will force AI companies to choose between paying for licensing and attracting regulatory and platform backlash that could shrink addressable markets overnight.

Artists are losing ground in a market that is simultaneously booming for AI. UNESCO’s fourth Re|Shaping Policies for Creativity report documents that digital revenues now represent 35 percent of creators’ income, up from 17 percent in 2018, and warns that generative AI outputs could drive revenue losses of 24 percent for music creators and 21 percent for audiovisual professionals by 2028. (unesco.org)

Why this matters to AI builders now
The numbers are a business design problem for model makers and platforms. If creators’ works are feeding training corpora without compensation and AI outputs begin to substitute human-made content at scale, then the economics of model deployment shift from product innovation to liability and licensing. That move changes margins, investor returns and go to market strategies for startups and incumbents alike. The UNESCO warning effectively converts a reputational risk into a quantifiable financial exposure that boards will want modelled in three scenarios: defensive, cooperative and regulatory. (unesco.org)

A second, independent signal comes from creators’ own industry bodies. CISAC has pushed a PMP Strategy study into the public conversation that reaches similar conclusions: under current regulatory arrangements, generative AI could put roughly 24 percent of music creators’ revenues at risk by 2028, producing a multi billion euro transfer of value away from authors and toward AI services. That is not abstract academic math; it is a redistribution that will change royalty pools and platform economics in measurable ways. (cisac.org)

What the market is already doing
Streaming platforms and distributors are not waiting for lawmakers. Deezer and others have developed AI detection and tagging systems after a flood of synthetic uploads, with estimates of tens of thousands of AI tracks arriving daily and a large share of plays flagged as fraudulent. Platforms are experimenting with removal, de‑monetization and labels to shield creators while preserving legitimate AI use. For AI companies, the lesson is blunt: product features that ignore provenance turn into trust liabilities, and trust liabilities kill network effects faster than most engineers can debug. (musicradar.com)

The legal and competitive squeeze
Legal fights are multiplying and the public narrative is shifting from novelty to compensation. Tech press and legal analysts note mounting lawsuits and a policy push to tie model training to licensing or to require transparency in datasets. That environment creates three commercial imperatives for AI firms: build rights clearance into data pipelines, price in licensing costs to unit economics, and design visible attribution so content ecosystems can measure substitution. Failure to do so risks losing access to key channels and partnerships with labels, publishers and studios. (decrypt.co)

If AI companies treat creative input as free fuel, the creative economy will treat that choice like a regulatory invitation.

The cost nobody is calculating
Consider a simplified scenario drawn from the CISAC estimates. If AI music outputs capture 20 percent of streaming revenues and the global streaming royalty pool is treated as a fixed pie, that 20 percent becomes a recurring expense or a revenue transfer. For an AI provider projecting 100 million in user revenue tied to music outputs, even a 5 percent licensing obligation or voluntary revenue share could swing profit margins by millions per year. The math is not dramatic in isolation but scales quickly as adoption grows and as platforms add detection and remediation costs. The polite term is operational risk. The blunt term is margin compression. Try telling investors that growth is cheaper than compliance; they will smile and mark the risk down anyway.

Practical implications for AI product teams
AI firms need defensible provenance and monetization pathways before market penetration becomes litigation bait. Product roadmaps should include dataset audits, opt in licensing partnerships with collecting societies, and UX for creator consent and attribution. On the engineering side, there is a near term tradeoff: faster iteration with raw scraped data versus slower, licensed training with clearer go to market. One option is white label enterprise licensing for library uses while restricting public streaming deployments until rights are settled. Yes, this sounds boring; it is also how regulated industries survive.

Technical and platform responses that matter
Detection systems, watermarking and content provenance will become standard components of any music or video generation stack. Platforms that invest early in robust detection and transparent revenue routing will have better leverage in policy conversations and in negotiating licenses. The alternative is a two tier market in which only well capitalized players can afford the compliance overhead while smaller companies either take on legal risk or lose access. That is not innovation-friendly, but it is realistic.

Risks and open questions that stress test the claims
Projections assume current regulatory regimes persist and that AI content will substitute for human work at projected scale. Both assumptions could be wrong if new licensing frameworks emerge or if consumer demand favors human authenticity. Another variable is fraud. If bots continue to inflate play counts, platforms may overcorrect and suppress legitimate AI uses, creating false negatives. There is also a geopolitical dimension: countries with limited digital skills and weak protections for creators may see faster displacement, while richer markets tighten controls.

Forward-looking close
The UNESCO findings are a market signal to the AI industry: treat creative content as a scarce, monetizable input and design for fairness now, or incur higher compliance and reputational costs later.

Key Takeaways

• UNESCO and industry studies project up to 24 percent revenue risk for music creators by 2028, translating into material market redistribution.
• AI companies must bake licensing, provenance and attribution into product economics to avoid sudden margin shocks.
• Platforms that deploy detection and transparent revenue routing will win creator and regulator trust.
• Ignoring these shifts is a strategy that preserves short term growth and destroys long term access to cultural markets.

Frequently Asked Questions

How much could AI-related licensing cost my startup?
Licensing costs vary by territory and use case, but scenario modelling from industry studies suggests even modest percentage shares of streaming revenue can reduce margins meaningfully. Startups should model a 3 to 10 percent range of revenue allocation for rights cleaning and legal protections in early financial plans.

Will labeling AI content solve the creators’ income problem?
Labeling increases transparency and may reduce substitution effects, but it does not by itself reallocate royalties or compel compensation for training data. Labeling is a necessary but not sufficient policy lever.

Can small AI companies compete if licensing becomes mandatory?
Yes, but not easily without cooperation. Aggregator licensing, collective bargaining with collecting societies and modular compliance tools will be essential to keep costs proportional to company size.

Are detection tools reliable enough to base policy on?
Detection has improved but is not perfect; adversarial inputs and hybrid human plus synthetic works complicate signals. Detection reduces harm but must be paired with governance and dispute resolution.

What immediate steps should AI platforms take this quarter?
Conduct a dataset audit, open talks with collecting societies, pilot provenance tags, and add legal risk to board reporting. Those steps buy time and negotiating credibility.

Related Coverage

Readers interested in the mechanics behind these shifts may want to explore how content provenance protocols are being built into media stacks, the evolution of collective licensing for datasets, and the economics of streaming pools in the age of synthetic content. Coverage that follows the policy decisions of major streaming platforms and collecting societies will be especially relevant to product and legal teams.

SOURCES: https://www.unesco.org/en/reshaping-creativity-reports https://www.cisac.org/Newsroom/news-releases/cisacs-2025-annual-report-spotlights-ai-advocacy-anti-fraud-frameworks-and https://decrypt.co/358535/ai-disruption-creator-earnings-unesco https://www.musicradar.com/news/slop-of-the-pops-over-30000-ai-generated-tracks-are-being-uploaded-to-deezer-every-single-day https://musically.com/2024/12/04/cisac-study-ai-music-outputs-will-be-worth-e16bn-in-2028/