AI Pledge

AI music has a real problem that platforms have been quiet about: every AI stream takes money out of the same payout pool that human artists are paid from. We are not going to do that here.

XECHO's commitment, in one sentence: AI activity on this platform cannot dilute a human artist's streaming income — and the math is verifiable on-chain.

1. Separate Revenue Pools, Three Rails, No Bridge

Human-artist streaming and AI-generated streaming are paid from different pools. An AI stream pays AI creators (and any opt-in artists in its training set). It does not subtract a cent from the human streaming pool.

The Human Pool is funded by base subscriptions, downloads, tips, and wave purchases — exactly as it is today. AI revenue flows through three deliberately separate rails, and they never cross through xecho custody:

• AI Tier Stripe rail. When a listener subscribes to the AI Tier, their card payment goes to Stripe. At end of each month, 80% of net subscription revenue is paid directly to AI creators via Stripe Connect transfers — proportional to verified listening that month. The platform retains 20% for infrastructure, moderation, AI-safety review, and generation compute subsidies. Money sits in xecho's Stripe platform balance only as long as it takes to settle (currently: collected by end of month, transferred to creators by the 1st of the following month). No platform-held wallet of stable, no fiat-to-crypto bridge, no custody point between the listener's card and the creator's Stripe account. Stripe's Connect destination scoping is the technical guarantee here: xecho can only transfer to Connect accounts the creator themselves connected.
• On-chain AI Pool rail. The XECHOAIPool contract is funded exclusively by on-chain activity: AI track tips (paid in BNB / USDT / USDC / XDRIP directly to the contract by the tipper), AI track downloads (XECHODownloads → AIPool routing at purchase time), AI wave/pack purchases (XECHOWavePayments, XECHOCuratedPacks → AIPool), and per-generation/per-distribution fees from the in-platform AI Studio (studio in development; the contract path is live on testnet today). Every cent in the AI Pool can be traced to a specific on-chain transaction. The xecho team never holds this money — the contract is the custodian, and the split is enforced in code, not policy. It is currently 2% reserve / 15% platform / 100% AI creator / 0% training pool. The training pool is the share that pays artists whose work trained a model, and it stays at 0% until we can say which artists a given generation drew on — paying it out before then would mean guessing, and a guess is not consent. Enrolling in the training pool today records that consent so you are included when the share is raised; it does not pay yet, and we will not pretend otherwise. The bounds the contract will not exceed regardless of configuration: reserve at most 5%, platform fee at most 25%, and AI creator plus training always summing to 100% of what remains. The live numbers are readable on /transparency, straight from the chain — if this paragraph and that page ever disagree, the chain is right.
• Voice license rail. Per-use voice-license payments go directly from the AI creator to the licensing artist at generation time, atomically with the AI output, via the XECHOVoiceLicense contract. Platform fee 5% (hard-capped on-chain at 10%). The licensing artist gets paid before the AI track is allowed to publish.

A listener on the base subscription who happens to play an AI track still pays 100% of their subscription fee into the Human Pool. AI listening does not redirect their dollars. The AI track is paid for through the AI Tier rail (separate subscribers), or at generation time on-chain, or via the on-chain per-stream rails — none of which deduct from the Human Pool.

This is the core complaint working artists have made about every other streaming service: AI tracks are mathematically taking food off their tables, today, in pools they never agreed to share. We answer that head-on with structural separation — on-chain rails are verifiable on /transparency; Stripe-rail settlements are published as monthly summary entries with per-period totals and transfer counts.

1.1 Why no bridge between Stripe and on-chain

The most common architecture for “Stripe subscription that pays into an on-chain pool” involves a custodial fiat-on-ramp service: the platform collects USD via Stripe, buys USDC at scheduled intervals, and deposits the USDC into the on-chain pool. This is technically straightforward and we explicitly refuse to build it. Custody points between rails are the single highest-loss category in crypto history — every cross-rail bridge has been drained or credentialed-into at some point. We do not want xecho to be the operator of one, ever.

The two-rail design is the consequence of that refusal. Stripe revenue settles via Stripe Connect. On-chain revenue settles via smart contracts. Neither rail pretends to be the other. The marketing claim “your AI Tier subscription funds AI creators” is true because Stripe Connect makes it true — we can prove it by showing you every Connect transfer in our Stripe dashboard and on the public monthly settlement page. We do not claim that Stripe revenue funds the on-chain AI Pool, because it does not, and pretending otherwise would require the custody point we refuse to operate.

1.2 What is out of scope

The pool separation governs streaming and subscription revenue. Two adjacent payment paths are deliberately outside it:

• Wave / license sales on wave-shop. A wave purchase is a work-for-hire payment from the buyer directly to the creator at sale time, not a stream royalty. AI-generated waves are priced and sold by their creators the same way human-creator waves are. The buyer pays the creator for a finished work; no AI Pool deposit happens because nobody is taking a cut of streaming revenue.
• Fan subscriptions to individual AI artists. A fan-sub to an AI artist routes via Stripe Connect direct-to-creator, same as a fan-sub to a human artist, with the platform application-fee taken at the moment of the charge. No AI Pool routing happens because the fan-sub is a creator-to-fan direct relationship, not a streaming pool allocation.

2. Human-First Discovery

AI-generated tracks live on a separate discovery surface. They do not compete with human artists in core feeds, autoplay, daily mixes, mood playlists, or editorial placement. Listeners who want AI music opt into it deliberately.

AI personas are not eligible for verified-artist status. AI tracks cannot be uploaded under a human artist's profile, ever.

3. Consent Before Training

We do not train AI models on copyrighted music we have not licensed. Every model available on XECHO is built from one of:

• Production music libraries we have licensed in writing
• Public-domain catalog
• A creator's own uploaded catalog, where they have warranted ownership
• An opt-in training pool with a posted revenue-share formula

Opt-in means opt-in. We do not auto-enroll uploads. We do not bury consent in a terms-of-service update. Today, the default state for every creator is non-enrollment — no creator's catalog has been added to any training pool, and there is no mechanism by which it could happen without an explicit, recorded consent. The in-dashboard toggle to opt in (and to revoke previously granted consent) is in development and will ship before any third-party model training is offered through XECHO. Until then: no enrollment exists, no enrollment can be created, no track on this platform is in any AI training set.

4. No Unauthorized Voice Clones

A voice or likeness model of any artist can only be created with that artist's explicit, written consent. They set the price. They approve each use. They can revoke access at any time, with payouts already earned honored.

5. Disclosure on Every AI Output

Every AI-generated or AI-assisted track on XECHO ships with:

• C2PA Content Credentials embedded in the file — cryptographically signed authorship and AI-involvement manifest
• DDEX AI metadata flags in the standard music industry feed
• A clear "AI-generated" or "AI-assisted" badge anywhere the track is presented to a listener

No hidden AI. No fake bands with photo-AI press shots posing as human acts. If it was made with AI, it says so — on the file, in the metadata, on the page.

5.1 The three classifications (binding under the Creator Agreement)

Every track, audiobook, Wave, and other Creator Content is classified as exactly one of three values in the ai_disclosure field at upload. This selection is recorded with the Content and is contractually binding.

• Human — Content created without material involvement of a generative AI model. Use of conventional production tools (DAW, plugins, AI-assisted mixing utilities, AI-assisted stem separation, traditional MIDI quantization, autotune) does not, by itself, render Content non-Human. What matters is whether the principal creative work was performed by a human author or by an AI model.
• AI-Assisted — A human is the principal creative author, but a generative AI model contributed substantially as a tool (e.g. AI-suggested chord progressions adopted with significant revision; AI-generated melodic ideas selected and edited by the human author; AI-generated lyric drafts heavily revised). The output reflects the human author’s creative judgment as the primary force.
• AI-Generated — The principal creative output was generated by an AI model and the human contribution is curatorial or directorial (prompt selection, model selection, post-processing, mixing of generated stems). Includes tracks fully composed by Suno, Udio, MusicGen, Stable Audio, or comparable text-to-music models; AI-generated vocal tracks using voice-cloning or text-to-speech models; mostly-AI tracks with light human editing.

The threshold between AI-Assisted and AI-Generated is a good-faith judgment by the Creator. Where the AI’s contribution exceeds what a human author could in good conscience claim authorship over, the correct classification is AI-Generated. False classification of AI status is a material breach of the Creator Agreement and grounds for immediate Content removal, royalty forfeiture, and Account termination.

5.2 Voice cloning and digital replicas — consent regime

Use of an AI model that replicates an identifiable person’s voice or likeness requires a signed, dated, and currently-valid release from the person whose voice or likeness is replicated, AI-Generated classification on the Content, and (where the use is sale-of-Wave) a per-use voice license arrangement administered through the XECHO Voice License contract. Voice cloning of public figures without explicit, current consent is prohibited regardless of jurisdiction. Statutory consequences include the Tennessee ELVIS Act (Tenn. Code § 47-25-1101 et seq.), California Civil Code § 3344 (right of publicity), California AB 2602 (digital replicas in performances), and California AB 1836 (post-mortem digital replicas), in addition to platform consequences.

5.3 Statutory and case-law alignment

This disclosure regime is aligned with:

• U.S. copyright — human-authorship requirement. 17 U.S.C. § 102(a) requires human authorship for copyright; Thaler v. Perlmutter, 689 F. Supp. 3d 1 (D.D.C. 2023), held that purely AI-generated works are not eligible for U.S. copyright registration. We do not represent that AI-Generated Content is copyrightable, and Creators of AI-Generated Content should understand the limited scope of any enforceable rights they hold in such Content.
• U.S. Copyright Office 2023 guidance on AI-assisted works applies to AI-Assisted Content with substantial human authorial contribution; the human portions may be registrable while purely-AI elements are disclaimed.
• EU AI Act, Regulation (EU) 2024/1689. Article 50 (transparency for AI interaction and AI-generated content) and Article 5 (prohibited practices) apply. The disclosure tools, badges, and AI-routed surfaces of XECHO meet the Article 50 transparency obligations effective 2026-08-02.
• Colorado AI Act, SB24-205 (effective 2026-02-01). XECHO is a Colorado entity; the impact-assessment, notice, and consumer-rights obligations of that statute apply to any feature meeting the “high-risk” threshold.

5.4 Detection + publish-gate enforcement (technical)

Disclosure is enforced contractually today and increasingly enforced technically as the detection pipeline matures. The current publish-gate is a database trigger (enforce_ai_track_disclosure_before_publish) that refuses to flip an AI track to public unless the disclosure-pipeline status is complete; this catches the case where a Creator marks a track AI-Generated at upload but skips the downstream disclosure steps. An automated AI-detection model that scores every upload 0–100 and surfaces discrepancies for human review is in development; today, mismatches between Creator-declared disclosure and the actual content are caught by listener reports and dispute resolution rather than by automated detection. Creators may submit a dispute through the Creator dashboard for human review at any time.

6. Verifiable Payouts — The Math

6.1 On-chain AI Pool split

When an AI track is tipped, downloaded, purchased as a wave, or settled as part of an AI pack, the XECHOAIPool contract receives the payment on-chain and immediately applies the following split. The rules below are not promises — they are inspectable in the contract source:

• 2% reserve off the top — funds refunds, dispute resolution, and EU AI Act / NO FAKES compliance costs
• 15% platform fee on the remainder — hard-capped on-chain at 25% maximum
• AI creator share — paid to the user who generated the AI track. On-chain today: 100% of the remainder after the platform fee. Target once attribution ships: 70%.
• Training pool share — split across opt-in training contributors per the attribution vector recorded at generation time. On-chain today: 0%, because no AI training pipeline exists on XECHO yet and no attribution rows can be recorded. Target once the attribution pipeline + training opt-in UI ship: 30%. This split is changeable on-chain by the protocol admin within bounded ranges, and the change will be announced in writing before it takes effect.

6.2 AI Tier Stripe-rail split

The Stripe AI Tier subscription rail is structurally different from the on-chain rail because Stripe Connect can split a single subscriber's charge cleanly between exactly two destinations, not the four-way split the on-chain pool supports. The Stripe rail therefore uses a simplified split:

• 80% creator pool — distributed monthly across all AI creators who received verified streams that month, proportional to their stream weight
• 20% platform fee — covers infrastructure, content moderation, AI-safety review, generation compute subsidies, and the small per-transfer overhead Stripe Connect charges

The training pool concept has no Stripe-rail allocation because the training pool is an on-chain mechanic and we do not bridge between rails. Training contributors are compensated entirely from on-chain inflows (per-generation fees, tips, downloads, wave/pack purchases). A creator whose work is exclusively in training contribution and never receives direct streams would be paid only via the on-chain rail, not the Stripe rail.

The monthly Stripe settlement is published as a public ledger entry containing the period's total collected revenue, the creator pool, the platform pool, the per-artist stream weight, and the Stripe transfer IDs for each payout. Every per-artist transfer is reproducible against Stripe's own dashboard records.

6.3 Voice-license payments

Voice-license royalties are not taken from either pool above. They are paid directly to the licensing artist at generation time through a separate contract — XECHOVoiceLicense — the artist's wallet receives the payment before the AI track is allowed to publish. Default platform fee on voice-license payments is 5%, hard-capped on-chain at 10%, so the licensing artist's take is structurally guaranteed to be at least 90%. The contract is deployed and live on testnet. The in-app flow that calls payForUse at generation time ships with the XECHO AI Studio — until then, no platform-hosted voice generation is offered and no voice clone licensing occurs through this contract.

6.4 What we do not claim

We do not claim that the on-chain AI Pool number you see on /transparency includes any Stripe-rail subscription revenue. It does not. The on-chain number reflects on-chain inflows; the Stripe rail has its own settlement ledger with its own number. The two are added in the marketing summary as “total AI revenue paid to creators” only when we are explicit that the figure combines both rails. Anywhere we display a single number, we label which rail it represents.

7. What We Will Not Build

The following are not on our roadmap, and will not be — at any price:

• Generative models trained on scraped, unlicensed copyrighted music
• Voice clones of artists who have not opted in
• AI tracks placed under a human artist's name or profile
• AI-seeded mood/focus playlists where listeners cannot tell the difference
• Algorithmic discovery that prefers AI tracks because they are cheaper to produce
• Any feature whose business case requires diluting human artist payouts
• A fiat-to-crypto bridge or operator-held custody account that moves Stripe subscription revenue into an on-chain pool. Bridges are the highest-loss infrastructure in this industry. The Stripe rail and on-chain rail are kept structurally separate to avoid being the custody point.

8. How to Opt Out

Every creator on XECHO can, at any time:

• Refuse to enroll any track in the AI training pool — this is the default state for every creator today, and the only state currently possible until the opt-in UI ships
• Withdraw previously enrolled tracks — future AI generations will no longer use them (enabled by the same in-development opt-in UI; no enrollment exists to withdraw today)
• Set an explicit Do Not Train flag broadcast in DDEX metadata to other platforms (in development — pending DDEX export pipeline)
• Decline to make a voice or style license available, with no penalty to discovery, payouts, or platform standing — declination is the default state today because no in-platform voice generation surface yet exists

Opting out of AI features does not deprioritize you anywhere on the platform. We do not punish artists for being human.

9. Hold Us To It

This pledge is the standard we are asking working artists to judge us by. If we ship something that breaks it, we want to hear about it loudly and publicly.

This pledge will evolve as the legal landscape, technical standards (C2PA, DDEX), and artist consensus evolve. Material changes will be announced before they take effect, not after.