Athlete portrait rights in Japan sit at the intersection of personal consent, club ownership, commercial licensing, and cross-border media distribution. Today these rights are managed through spreadsheets, fax-based NDA workflows, and bilateral email chains. No single system tracks who has consented to what, which territories are licensed, or whether a campaign has actually run.
When an athlete transfers between clubs, the cascade of consent revocations, license suspensions, and settlement freezes is handled manually — often weeks after the transfer is complete. Infringement detection is reactive. Revenue settlement is slow and opaque.
IP Bank RWA replaces this with a single sovereign ledger: every consent, every license, every settlement instruction anchored to a cryptographic proof on Midnight — visible to the rights holder, auditable by the regulator, and invisible to everyone else.
The RWA instance is configured for high-sensitivity, low-autonomy operation. Every AI decision that touches consent, settlement, or disclosure requires human approval. The Midnight blockchain operates in PRIMARY mode — full Compact smart contracts, zero-knowledge proofs, and shielded settlement.
Foundation-AI is the AI operating system that powers every intelligent behaviour in the IP Bank. It is not a chatbot bolted onto a database. It is a multi-agent architecture where specialised AI systems — Triton, Cortex, the Scout Fleet, and the Aristotle domain engine — operate as a coordinated organism, each with defined capabilities, strict autonomy boundaries, and a shared nervous system called LogOS.
Triton is the user-facing intelligence layer. It receives every query — from a rights holder asking about their revenue, to a licensee searching for available IP, to an operator checking system health — and routes it through one of four specialised lanes: inquiry, rights search, operations, or workflow. Each lane carries its own Semantic Airgap configuration. An inquiry about public market trends operates under PUBLIC_RESEARCH_ONLY. A workflow action involving a license grant operates under DENY — no external model touches the data.
Triton maintains cross-session collaboration memory: if a negotiation spans multiple conversations over weeks, Triton retains the full context without re-ingesting documents. Its prompt evolution system means that Triton's effectiveness improves with use — successful interaction patterns are crystallised into reusable playbooks, promoted across agent groups, and deduplicated at a 0.75 similarity threshold to prevent redundancy.
Cortex monitors the entire platform through three operating modes. Reactive Red handles things that are broken now — auto-remediating Tier 2 actions like service restarts, settlement retries, and scout re-dispatches without human intervention. Proactive Amber detects things that will break soon — alerting operators and proposing remediation before failures cascade. Strategic Blue identifies how the platform gets stronger — suggesting configuration improvements, workflow optimisations, and policy refinements based on accumulated operational data.
Cortex operates under a four-tier action hierarchy. Tier 1 (observe) is unrestricted: monitor, correlate, simulate, propose. Tier 2 (reversible) is auto-executed: restart services, retry settlements, adjust thresholds. Tier 3 (approval-required) is proposed and held: schema migrations, policy changes, contract migrations wait for an operator to approve. Tier 4 (forbidden) is absolute: Cortex can never bypass the Semantic Airgap, weaken the audit trail, disable the circuit breaker, or reveal secrets. An independent watchdog — an unintelligent bash process running on a five-minute interval — kills Cortex entirely if it exceeds any threshold. The watchdog has no intelligence. It cannot be reasoned with. It is the final safety net.
LogOS is the connective tissue. It provides the event bus (asyncio queues in-process, Kafka across services), the memory system (episodic, semantic, hive, and graph), the agent registry (heartbeat-driven presence and capability discovery), and the deliberation protocol (multi-agent voting for high-blast-radius decisions). Every significant decision produces a DecisionTrace — a JSONL record of what was decided, why, what alternatives were considered, and what signals drove the outcome.
The cost tracker monitors ten distinct cost sources — from Triton's frontier model usage to Midnight gas fees to Kafka throughput — with EWMA-based anomaly detection that triggers warnings at 2x burn rate and critical alerts at 5x. For the RWA instance, where settlement values can be significant and a runaway agent could generate material cost exposure, this is not a reporting tool. It is a budget enforcement mechanism.
FMIE is the perception layer. It doesn't search — it sees. Every piece of media that enters the platform — a contract PDF, an athlete photograph, a campaign screenshot, a licensee's promotional video — passes through FMIE before it touches any other system. FMIE is the gatekeeper that determines what something is, whether it is authentic, and what rights attach to it.
When an athlete's portrait rights are registered, FMIE produces a Media Passport — a comprehensive sidecar artifact that travels with the asset throughout its entire lifecycle on the platform. The Media Passport is not a single document. It is a structured bundle of machine-readable attestations, each generated by a different FMIE capability, each independently verifiable.
The passport contains the perceptual fingerprint (a mathematical representation of the athlete's visual identity that survives cropping, colour shifts, style transfer, resolution changes, and partial occlusion), the authenticity score (FMIE's confidence that the original media is unmanipulated), the sensitivity tags (face detected, PII present, biographical context inferred), the disclosure transforms (silhouette for F0, redacted for F1, full for F3+), the language detection result, and the lineage graph (a directed acyclic graph tracing every derivative version of the asset back to the registered original).
The Media Passport hash is anchored to the Asset Contract on Midnight at the point of registration. This means the passport's contents are cryptographically bound to the on-chain asset — any tampering with the passport after registration is detectable by comparing the current hash against the anchored proof. When a licensee receives an asset via the CDN Gateway, the passport is delivered alongside it. When the Scout Fleet detects a potential infringement, it compares the suspect media's fingerprint against the passport's registered fingerprint. When a dispute arises, the passport's lineage graph and authenticity attestations form the evidentiary foundation.
Every piece of campaign evidence submitted by a licensee — screenshots of advertisements, recordings of broadcast spots, photographs of merchandise in-store — passes through FMIE's authentication pipeline before it is accepted as proof of campaign execution. The pipeline operates in three stages.
First, manipulation detection. FMIE analyses the media for signs of synthetic generation or post-capture alteration: inconsistent lighting and shadow geometry, compression artefact anomalies, frequency-domain signatures characteristic of GAN-generated imagery, splicing boundaries, and clone-stamp patterns. Each detected anomaly is localised to a specific region of the image or frame of the video, producing an authenticity regions map that shows exactly where and why FMIE's confidence drops.
Second, deepfake detection. For media containing faces — which is the majority of athlete portrait rights content — FMIE runs a dedicated face authentication pipeline. This compares the face in the submitted media against the registered fingerprint in the Media Passport, checking not just visual similarity but physiological consistency: skin texture continuity, iris reflection patterns, ear geometry, and temporal coherence across video frames. A deepfake that replaces an athlete's face with a synthetic version that looks visually similar will fail on temporal coherence — the micro-movements of a real face over consecutive frames follow patterns that current generation models cannot perfectly replicate.
Third, provenance verification. FMIE examines the media's metadata chain — EXIF data, encoding signatures, compression history — to determine whether the media has been re-encoded, screenshotted, or passed through a generative pipeline. A photograph taken with a camera and submitted directly will have an intact metadata chain. A photograph that has been passed through a face-swap model and re-saved will show a broken chain — even if the visual output is indistinguishable to a human eye.
The output of the authentication pipeline is a challenge bundle — a structured artifact containing the authenticity score, the anomaly region map, the deepfake analysis results, the provenance chain assessment, and FMIE's overall confidence rating. This bundle is stored in Alexandria as campaign evidence and referenced by the Scout Fleet's Campaign Verifier mission. If the authenticity score falls below the threshold, the campaign evidence is quarantined and the publisher is notified. The settlement associated with that campaign milestone is frozen until a human reviewer resolves the dispute.
The Media Passport and the deepfake detection pipeline work as a pair. The passport establishes what the authentic original looks like — its fingerprint, its provenance, its sensitivity profile. The authentication pipeline then measures every subsequent piece of media against that baseline. Together, they create a closed loop: the platform knows what is real because it registered the original, and it can prove what is fake because it can measure the distance from that original with mathematical precision.
For the RWA instance, FMIE's sensitivity tagging is critical. Athlete imagery carries inherent PII — faces, names, biographical context. FMIE tags sensitivity levels and generates disclosure transforms: versions of the asset that can be shared at different phases of the licensing workflow without exposing the full original. A buyer at F0 sees a silhouette and metadata. At F1 (post-NDA), they see the full asset. This graduated disclosure is enforced by the Asset Delivery Gateway and gated by the Disclosure Contract on Midnight, not by trust.
Foundation-AI's defining principle for the RWA instance is low autonomy, high intelligence. The agents are extraordinarily capable — but every action that touches consent, settlement, or disclosure requires a human to approve. The AI proposes. The human disposes. The blockchain proves.
Midnight is a privacy-preserving blockchain purpose-built for regulated industries. In the RWA instance, it operates in PRIMARY mode — the most comprehensive integration available. This means every significant state change in the platform is not merely logged but cryptographically anchored with zero-knowledge proofs that can be verified by authorised parties without revealing the underlying data.
A talent agency needs to prove to a regulator that an athlete's consent was obtained before a license was granted — without revealing the commercial terms of the license. A licensee needs to prove to a brand partner that they hold a valid territory license — without disclosing the royalty rate. An investor holding IP Tokens needs to prove their holding is within the jurisdictional cap — without revealing the total number of tokens in circulation. Zero-knowledge proofs make all of these possible. The proof is mathematically verifiable. The data behind it remains shielded.
Midnight uses a contract language called Compact, purpose-built for privacy-preserving business logic. In the RWA instance, Compact contracts are deployed across the full lifecycle:
When a royalty payment is settled through PayPay, the Settlement Service generates a settlement instruction that is anchored to Midnight as a shielded receipt. The receipt proves that a payment of a specific amount was made, from a specific buyer, to a specific rights holder, at a specific timestamp — but the amount, the parties, and the terms are all shielded. Only the rights holder (via their viewing key) and the regulator (via a separate regulatory viewing key) can decrypt the full details. Everyone else sees a valid proof and nothing more.
This is not encryption. Encryption hides data and requires trust in the key holder. Shielded settlement proves facts about data without revealing the data itself. The difference matters in regulated markets: a regulator can verify compliance without requiring the platform to expose commercially sensitive information to a third party.
Every consent record — the athlete's agreement to license their portrait rights under specific conditions — is anchored to Midnight at the moment of creation. When consent is revoked, the eight-step Consent Cascade Orchestrator executes its saga and anchors the revocation as a new proof that references the original consent proof. This creates an immutable chain: consent granted → licenses issued → settlements made → consent revoked → licenses suspended → settlements frozen. The entire chain is verifiable. No single step can be forged, backdated, or deleted.
For athlete transfer scenarios, the Athlete Transfer Engine triggers a consent cascade that revokes all consents under the prior club and anchors the entire transfer as a single atomic Midnight proof. The new club's onboarding creates a fresh consent chain that is explicitly disconnected from the prior one — there is no inheritance of consent across club boundaries.
The Scout Fleet runs a dedicated Midnight Anchor Verifier mission every four hours in AUGMENTED mode. This mission compares the platform's database state against the Midnight chain — detecting any drift between what the platform believes to be true and what the blockchain can prove. If the database shows a license as active but Midnight shows the corresponding consent as revoked, the verifier flags the discrepancy as critical. This is not a nice-to-have monitoring feature. It is the mechanism that ensures the platform's operational state never diverges from its cryptographic truth.
Midnight's role in the RWA instance is not "blockchain for the sake of blockchain." It is the enforcement mechanism for a specific promise: that an athlete's consent is sovereign, that their commercial terms are private, and that the proof of both is stronger than the word of any intermediary.
Japanese athlete portrait rights carry NG (No Good) expression restrictions — specific poses, contexts, or product categories that a talent agency has declared off-limits. The NG Rule Engine evaluates every incoming license request against the rights holder's declared restrictions before it reaches a human reviewer. A merchandise request for an energy drink campaign is automatically flagged if the athlete's NG list prohibits alcohol-adjacent branding.
When an athlete changes club, the platform triggers a consent cascade: all existing consents under the prior club are revoked, active license grants are suspended, pending settlements are frozen, asset delivery tokens are invalidated, and all affected parties are notified in their preferred language. The entire cascade is anchored to Midnight as a single atomic proof. The new club's rights holder is then onboarded and existing licensees can re-request under the new consent regime.
Before any license is granted, the Exclusivity Engine checks for conflicts — territory overlaps, category exclusivity, temporal collisions. A licensee requesting exclusive merchandise rights for Kanto cannot be granted if another licensee already holds non-exclusive rights for the same territory and category with a wider date range.
An eight-step idempotent saga that handles consent revocation across every dependent system — from license suspension through to Midnight proof anchoring. Each step is individually retriable, and the entire cascade can be replayed from any failure point without side effects.
The platform supports both a fast-path flow for pre-approved templates and a full five-phase negotiation (F0 through F4) for bespoke deals.
Royalty settlement runs on PayPay rails with JPY as the primary currency. The Settlement Service calculates royalties automatically based on pre-configured rules, locks FX rates at the point of instruction, and retries failed settlements with exponential backoff. Every settled payment is anchored to Midnight as a shielded receipt — the amount is provable to the rights holder without being visible to third parties.
IP Tokens are not database entries with a blockchain receipt. They are native Midnight assets. The entire token lifecycle — minting, offering, purchasing, distributing revenue, and redeeming — executes as Compact smart contract transactions on the Midnight chain. The platform's Tokenization Engine orchestrates these transactions, but the source of truth is the Token Contract on Midnight, not the platform's database.
This distinction matters. A token that exists only in a database can be modified by the database administrator. A token that exists on Midnight can only be modified by a valid Compact transaction signed by an authorised party. The platform cannot unilaterally alter holdings, redistribute revenue, or create tokens that violate the contract's minting rules — even if the platform's own database is compromised.
A talent agency registers an athlete's portrait rights as an Asset Contract on Midnight. The agency then deploys a Token Contract as a child of that Asset Contract, defining the total supply, the minimum and maximum holding per wallet, the jurisdictional gating rules, and the revenue distribution formula. The total supply represents fractional economic interest in the licensing revenue generated by that specific asset.
Tokens are offered to qualified investors (verified via the Buyer Contract's KYC proof). Each purchase is an atomic Compact transaction: the buyer's payment is confirmed on PayPay, the Token Contract updates the holding ledger, a holding proof is generated, and the settlement receipt is shielded — all within a single Midnight transaction. There is no window between payment and token allocation where the state is inconsistent.
Revenue distribution happens on a configurable schedule. When royalties are settled against the parent Asset Contract, the Token Contract's distribution function calculates each holder's pro-rata share based on their holding at the distribution snapshot. Distribution instructions are pushed to PayPay for settlement, and each payout is anchored as a shielded receipt. A holder can verify that they received their correct share using their viewing key — without seeing any other holder's payout.
Token redemption converts a holding back to settlement currency. The holder submits a redemption request, the Token Contract validates it against the redemption policy (minimum holding period, redemption window, remaining supply floor), and if valid, triggers a PayPay settlement instruction. The redeemed tokens are burned on-chain — reducing total supply and proportionally increasing the remaining holders' share of future distributions.
If the parent Asset Contract transitions to SUSPENDED (e.g. during an athlete transfer), the Token Contract enters a frozen state: all operations halt, distributions are deferred, and a notification is dispatched to all holders. When the asset is reactivated under a new consent regime, the Token Contract resumes — but any deferred distributions are recalculated against the new consent terms, not the old ones.
| Capability | Implementation |
|---|---|
| Settlement Rails | PayPay JPY — QR payments, eKYC, P2P transfers, bank settlement |
| FX Handling | Rate locked at instruction creation, multi-currency support |
| Token Lifecycle | Midnight Compact contracts — mint, purchase, distribute, redeem, burn |
| Fractionalisation | Configurable supply, per-wallet caps, jurisdiction gating, pro-rata distribution |
| Proof Anchoring | Midnight PRIMARY — every token operation is a ZK-proven on-chain transaction |
| Parent-Child Binding | Token Contract inherits state from Asset Contract — suspension cascades automatically |
| Compliance | Holding cap enforcement, jurisdiction gating, consent-freeze cascade, redemption policy |
The Scout Fleet operates thirteen mission types, running continuously. For the RWA instance, the most critical are the Usage Monitor (crawling web and social media for unauthorised use of athlete imagery), the Campaign Verifier (confirming that licensees are actually running campaigns they've paid for), and the Cross-Border Monitor (detecting territory violations when Japanese-licensed content appears in unlicensed markets).
Scouts operate in three execution modes. LOCAL missions use only the on-platform media intelligence engine at zero model cost. AUGMENTED missions add a frontier language model for reasoning. AUTONOMOUS missions can spawn sub-scouts to pursue leads. All findings feed into the Hive — a peer-to-peer mesh that lets any scout's discovery benefit every other scout.
The Semantic Airgap prevents any confidential rights data from leaking to external AI models. In DENY mode (the default for RWA), no external model call is permitted. When Triton needs frontier reasoning for a negotiation assist, it operates in ABSTRACT_REASONING_ONLY mode — sending surrogated terms rather than raw contract data.
All personal data handling complies with APPI (Japan's Act on the Protection of Personal Information) and GDPR for cross-border scenarios. The erasure protocol replaces content with [ERASED] markers, zeroes embeddings, and retains only an audit skeleton — the proof that something existed without revealing what it was.
The consent grace period for the RWA instance is zero days — revocation is immediate and irrevocable. This reflects the sensitivity of athlete portrait rights, where any delay between revocation and enforcement creates legal exposure.
The following traces the complete lifecycle — from a talent agency's first interaction with the platform through to an investor receiving their first royalty distribution from a fractionalised athlete portrait rights token.
This is the complete picture: a talent agency onboards in a day, registers an asset in minutes, reaches a global marketplace immediately, negotiates with AI assistance under privacy constraints, settles royalties in real time, fractionalises the asset into tokens governed by on-chain smart contracts, distributes revenue to investors with cryptographic proof — and the entire system is defended by an autonomous fleet of monitoring agents that never sleep. Every step is anchored. Every proof is verifiable. Every consent is sovereign.