AI Ownership & Attribution Policy

§1. Scope and Key Definitions

This AI Ownership & Attribution Policy ("AI Policy") applies to any use of artificial intelligence, machine-learning or generative-model functionality made available by SynthVerse to creators via:

• SynthVerse Studio™ — a browser-based digital-audio workstation and creation suite that provides optional AI-assisted tools including lyric suggestion, melody completion, stem separation, mastering assistant, drum-loop synthesis, and mix-style transfer, all under the direct manual control of a human creator working in a traditional multitrack timeline.
• SynthVerse Forge™ — a dedicated AI-generation tool suite that provides higher-level generation including text-to-audio stems, text-to-music sketches, voice characterisation, style-based track expansion, and cover-art generation. Forge is expressly designed to be used with human curation, refinement and re-integration into Studio rather than as a one-click publishing tool.
• AI features in any other SynthVerse product including, where applicable, AI-translated lyric captions, automatic-content-description generation for playlists and releases, automatic genre and mood taggers, and Vault automated provenance classification.

Key definitions. For the purpose of this AI Policy:

• Human Creative Direction ("HCD"). The process by which a human creator exercises meaningful, non-trivial creative choice, including: defining an artistic concept or brief, selecting prompts with specificity and intent, choosing among multiple candidate generations, arranging, composing, ordering, mixing, mastering, overdubbing, re-synthesising stems, editing for artistic intent, selecting or modifying instrumentation, writing lyrics, applying structural or harmonic changes, and integrating AI-produced elements with entirely original human-produced elements into a coherent whole. HCD excludes mere "art direction by generic prompt" such as typing "make a pop song" with no further selection, refinement or editing.
• AI-Assisted Output. Any output produced using Studio or Forge where Human Creative Direction has been applied to the result before publication.
• Fully-Automated Output ("FAO"). Any output published or submitted to SynthVerse with zero or merely de minimis human curation, selection, editing or arrangement between the act of pressing "Generate" and the act of publishing. Examples include: publishing the first Forge generation unmodified, batch-generating 100 tracks and publishing them without individual listening or editing, publishing auto-generated cover art without selection or refinement, and prompt-generated lyrics pasted verbatim into a release without revision, performance direction or melodic arrangement by a human creator.
• Base Model. A large-scale pre-trained machine learning model made available by a third-party provider and integrated into Studio or Forge. Examples (at the time of policy publication) include: the music generation model family codenamed "Helix-1", lyric-assist "VerseLM-7B", stem-separation "SeparaNet-3", mastering-style-transfer "MasterKit-XL", and text-to-image "Canvas-SDXL". The actual list of base models is maintained in the public Forge model register and updated whenever a new base model is introduced.
• Provenance Grade. The SynthVerse A / B / C / D letter grade described in §3 that is displayed on every release detail page and embedded in audio-file ID3 tags and Vault certificates, communicating to listeners the degree of Human Creative Direction vs. Fully-Automated Output and the training-data provenance transparency of the base models used.

This AI Policy is incorporated by reference into the SynthVerse Terms of Service. Capitalised terms not defined herein have the meaning given to them in the Terms of Service.

§2. You Own What You Direct

The rule applies to Provenance Grades A, B and C (all of which, by definition, require more than de minimis HCD). The following rules of ownership, authorship and license apply, ranked by grade:

• Grade A — HUMAN-DIRECTED. Full, unconditional ownership by the human creator. SynthVerse asserts no copyright, authorship claim, residual license or royalty claw-back in Grade A output beyond the narrow operating license granted in the Terms of Service §4. Base-model providers receive no royalty, no attribution beyond the grade badge and base-model label, and no authorship or co-ownership claim against Grade A works — their models are treated as tools, analogous to a synthesiser, a drum machine or a software sampler used by a human producer.
• Grade B — MODEL-ASSISTED. Same as Grade A for ownership: full unconditional ownership by the human creator. The only differences from Grade A are: (a) the Provenance Grade badge signals to listeners that AI contributed more than a trivial fraction, and (b) the base-model attribution label is displayed with a "Model-assisted" tag. No third-party authorship or royalties accrue to any base-model provider.
• Grade C — MODEL-AUGMENTED. Same ownership as Grade A and B: the human creator retains full ownership of the integrated whole. However, because a significant structural component of the work (e.g., an entire verse of auto-generated lyrics, an entire generated drum part used verbatim, or a generated harmonic progression used unedited) was adopted with comparatively light curation, the creator certifies by selecting Grade C that they have reviewed the underlying base-model provider's output license and that their use of that specific verbatim component is permitted. SynthVerse treats the creator as owner; disputes between the creator and a base-model provider about verbatim appropriation of a large generated block are handled under §8 (Disputes).
• Grade D — FULLY-AUTOMATED OUTPUT ("FAO"). Grade D works (Fully-Automated Output published without meaningful HCD) receive a limited ownership treatment on SynthVerse, reflecting both emerging copyright case law in multiple jurisdictions and SynthVerse's commitment to genuine human creativity. Specifically:
  • The human creator retains ownership of the prompt, the curatorial decision to publish (if any), and the choice of publication metadata.
  • Copyright in the output audio/visual content itself — where and to the extent that such copyright arises under applicable law — is treated as held by the human creator on a rebuttable basis for the purpose of SynthVerse internal operation (royalty distribution, Vault listing, DSP delivery) only.
  • SynthVerse does not represent to third parties, listeners, DSP partners, collective management organisations or courts that Grade D output qualifies for copyright protection as an original work of authorship in every jurisdiction. It is the creator's sole responsibility to determine whether copyright protection is available for their FAO in their target jurisdictions, and to bear the risk of non-protection should a court or CMO decline protection.
  • Grade D releases are ineligible for Featured, Verified, Trending, New Releases front-page placement, chart ranking, SynthVerse Radio linear rotation, and Documentation Vault Immutable-grade certificates (they may receive Vault "Standard-grade" timestamps only).
  • Grade D releases are limited to a maximum of 5 per account per calendar month on Free and Basic plans, 20 per account per calendar month on Professional, and an audited, substantiated quota on Enterprise. This quota is designed to prevent flood-the-market spam of FAO on the platform while still allowing creators to publish raw demos and sketches.

No transfer of authorship from human to SynthVerse or to base-model provider. Under no circumstances does SynthVerse, any SynthVerse affiliate, or any base-model provider become the "author" or "co-author" of a work created using Studio or Forge. Authorship is always attributed to the human creator(s) on the release page and in downstream DSP metadata. Base-model providers receive, at most, the specific attribution label described in §5.

Moral rights. The human creator is treated as the holder of moral rights (attribution, integrity) for purposes of the South African Copyright Act, 1978 s.20 and analogous provisions in other jurisdictions, for all Grades A through D, except that integrity claims on Grade D works — which were not substantially shaped by human creative direction — are limited to cases of outright distortion or misattribution, not stylistic critique.

§3. SynthVerse Provenance Grades — A / B / C / D

Every release published on SynthVerse that uses any Studio or Forge AI feature must carry a Provenance Grade, self-assessed by the uploading creator and subject to random audit and automated detection by the Vault provenance classifier. The four grades are:

Grade self-selection and audit. The uploading creator selects the grade at release-submission time. SynthVerse runs an automated Vault provenance classifier on every AI-tagged release that analyses: (a) embedded Forge/Studio generation metadata hashes (which track number of candidate outputs rejected before selection, number of edits applied in Studio timeline, whether mastering was AI-assisted or manual); (b) audio embedding similarity to known base-model training-set fingerprints where disclosed; and (c) edit-distance metrics in the Studio timeline project file. Where the classifier's estimate is materially different from the creator's self-declared grade, Vault flags the release for a manual review by the AI Policy team, who may: (a) confirm the creator's grade, (b) adjust the grade one letter with a written explanation, or (c) escalate to Trust & Safety for a potential enforcement action under the Acceptable Use policy. Grade adjustment decisions are appealable under §8.

§4. Training-Data Provenance Disclosures

SynthVerse commits to the highest commercially-reasonable degree of training-data transparency for every base model integrated into Studio or Forge. Training-data provenance is the foundation of accountability, and of trust between creators, listeners and the platform.

Forge Model Register

SynthVerse maintains a public, versioned Forge Model Register at https://synthverse.co.za/legal/ai-register (or, for direct API users, via GET /v1/models) that lists, for every base model currently available in Studio or Forge:

1. The model's commercial name, version number, release date and SynthVerse integration date.
2. The identity of the model provider (legal entity name, country of domicile), and whether SynthVerse licensed the model under a commercial enterprise license, an open-source license (e.g., Apache 2.0, MIT, CreativeML Open RAIL-M), a custom negotiated license, or co-developed / fine-tuned the model in-house.
3. A plain-English summary of the categories of training data used to train the base model (e.g., "publicly-available audio recordings published before 2025, labelled public-domain or CC-BY", "libri-speech style public-domain audiobooks", "proprietary licensed sound-effects library", "MIDI datasets from Lakh", "user-contributed training data explicitly opted in to Forge training programme by the uploading creator").
4. The approximate size of the training dataset in tokens / audio-hours / parameters (within commercially-reasonable disclosure that does not compromise the provider's trade secrets).
5. A provenance classification applied by SynthVerse to the model's training sources: Level 1 (Vetted), Level 2 (Mixed), or Level 3 (Opaque), defined as follows:
   • Level 1 — Vetted. SynthVerse has received from the provider, and independently spot-audited, machine-readable inventory records listing at least > 95% of individual works in the training set, together with evidence of license or public-domain / CC0 status for each listed work, for every category of copyrighted material that forms > 1% of the training set. SynthVerse itself uses Level 1 models only in Forge's premium highest-quality generation modes.
   • Level 2 — Mixed. The provider has given SynthVerse summary-level categories and sources of training data, has represented that works were scraped or collected in reliance on fair use / fair dealing / text-and-data-mining exceptions, and has offered a takedown / opt-out mechanism for rightsholders who do not wish their works to be included in future training runs. SynthVerse does not have individual work-level inventories, but has a documented opt-out escalation procedure for rightsholders.
   • Level 3 — Opaque. The provider has declined to disclose the specific composition, sources or licensing basis of the training set beyond high-level generalities (e.g., "publicly-available data"). SynthVerse allows Level 3 models only for internal, non-production features such as developer-grade draft sketching on Forge's "Experimental" toggle, and their output is ineligible for release at Grades A, B or C — publication of output from a Level 3 model automatically caps the Provenance Grade at D.
6. The opt-out / takedown procedure for rightsholders who do not wish their works to be used for future training runs of that model (where offered), with a direct email or web-form link to the provider's opt-out mechanism.
7. Any documented known limitations, biases or safety-guardrail behaviours of the model (e.g., "this voice-synthesis model will refuse to generate the voice of a living public figure without separate consent evidence submitted to SynthVerse").
8. The applicable output license terms granted by the model provider to end-users of SynthVerse, summarised and — where commercially allowed — linked in full.

The Forge Model Register is updated no fewer than once per calendar quarter, and any material change to the training-data provenance classification (e.g., a model upgraded from Level 2 to Level 1 following an audit) is accompanied by a note in the register and an in-app banner to affected creators who have previously used the model in the prior 12 months, so they may — if they wish — re-submit their releases for an automated Provenance Grade re-evaluation at no cost.

Creator opt-out from training use of their own works

By default, works you upload to SynthVerse are NOT used to train SynthVerse's in-house base models or any third-party models that SynthVerse sub-licenses. You may optionally opt in to the Forge Training Contributor Programme from Account > Privacy > AI Training Opt-In, which grants SynthVerse and its contracted Level 1 model providers a non-exclusive, royalty-free license limited to training model parameters on your work, in exchange for a training-credit royalty pool share based on the minutes of your content consumed by training runs. If you opt in, you may opt out at any time with 30 calendar days' written notice to the DPO, and future training runs after the effective date will exclude your works (it is not technically feasible to "un-train" a model already completed, but the opt-out is effective for all future fine-tunes and new model versions).

§5. Base-Model Attribution and Release Labels

Every release that uses Forge or Studio AI at any level — including Grade A — carries visible, machine-readable and human-readable attribution information, so that listeners, DSPs, collective management organisations, journalists and Vault auditors can see exactly which models were used and at what Provenance Grade. Attribution is never a co-authorship claim; it is a provenance disclosure, analogous to credit on a record sleeve for "producer", "engineer", "drum machine model", or "studio where recorded".

Where attribution appears

• Release detail page (every Grade B, C, D release — Grade A optional on creator's request): A small "AI Provenance" section in the Credits block showing: (a) Provenance Grade badge (coloured letter with tooltip explanation); (b) list of base models used ("Helix-1 music-gen", "Canvas-SDXL cover-art", "VerseLM-7B lyrics draft"); and (c) model-training Level (1/2/3) with a link to the appropriate entry in the Forge Model Register.
• ID3 / metadata tags embedded in downloads (every Grade B/C/D release, Grade A if the creator opts in): TXXX "SYNTHVERSE_AI_GRADE" = A/B/C/D; TXXX "SYNTHVERSE_AI_MODELS" = comma-separated list; TXXX "SYNTHVERSE_AI_LEVEL" = 1/2/3; UFID "https://synthverse.co.za/vault/" = Vault certificate URN containing full AI provenance hash chain.
• Documentation Vault certificate (every Provenance Grade on request): The certificate reproduces the Grade, model list, training level, a hash of the Forge generation parameters used, a hash of the Studio edit timeline (if applicable), and — for Grade A — the HCD-time-to-AI-time ratio metric as a supporting exhibit.
• DSP delivery metadata (on DSPs that support custom provenance fields): Provenance Grade and model list are delivered as proprietary metadata fields on releases that have opted in to DSP metadata disclosure. Where DSPs do not yet support custom fields, the information is instead appended to the copyright notice line in the form "[Grade X | models: Y, Z | synthverse.co.za/legal/ai-policy]".
• Search, recommendation and discovery surfaces (every release): The one-letter Provenance Grade badge appears as a small coloured chip on release cards, playlist cards, chart entries and artist pages, so listeners can filter by Grade if they wish. Users can set Saved Filters in Account > Preferences (e.g., "Show me only Grade A and Grade B in recommendations").

Grade badge colour palette and design

To avoid ambiguity the four Provenance Grade badges use a consistent colour and iconography system throughout SynthVerse:

• Grade A · deep emerald green chip with a check-mark icon and the label "A · Human-Directed". Tooltip text: "Human-curated throughout. AI used as a tool, not as a creator."
• Grade B · sky-blue chip with a pencil icon and the label "B · Model-Assisted". Tooltip: "AI generated initial drafts, human heavily curated and edited. See credits for models used."
• Grade C · amber chip with a sparkles icon and the label "C · Model-Augmented". Tooltip: "Large AI-generated blocks used with moderate curation. See credits for models and license notes."
• Grade D · slate-grey chip with a robot icon and the label "D · Fully-Automated". Tooltip: "Published with minimal or no human curation. Copyright eligibility varies by jurisdiction."

Creators may not override their release's Provenance Grade badge to display a higher grade than the Vault classifier assigns after audit. They may, however, voluntarily downgrade their declared grade (e.g., a creator who self-assessed Grade A and who was assigned Grade A can choose to display a Grade B badge for transparency). Upward grade reassignment is only possible after a successful Grade Appeal (§8).

§6. SynthVerse Studio vs. SynthVerse Forge — Rule Differences

Studio and Forge serve different use cases and are therefore subject to slightly different default assumptions and minimum curation requirements. The differences are summarised in the table below, and supplement the general rules in §2–§5.

§7. Third-Party Model Providers and Their Terms

Where Studio or Forge integrates a base model owned, trained or distributed by a third-party provider, creators' use of that model is additionally governed by that provider's end-user output license, acceptable-use policy and safety guardrails — to the extent those terms are enforceable and do not conflict with this AI Policy or the SynthVerse Terms of Service. SynthVerse passes through to the creator the benefit of any commercial output-usage rights the provider grants, and the creator agrees to comply with the provider's posted acceptable-use restrictions (for example, prohibitions on generating child sexual abuse material, deepfakes of non-consensual sexual content, targeted incitement to violence, or regulated financial or medical advice).

SynthVerse pre-vets every base model provider before integration. Minimum requirements for a provider to be listed in the Forge Model Register at Levels 1 or 2 include, without limitation:

1. A commercial enterprise license (or equivalent open-RAIL license) explicitly granting end-users worldwide, perpetual, non-revocable, royalty-free commercial-use rights to output created by the model using the end-user's own prompts, free of per-unit royalties beyond the Forge plan subscription price paid to SynthVerse.
2. Representations and warranties from the provider that: (a) the training process was lawful under at least one of: licensed data, public-domain data, CC0/CC-BY data with attribution, training conducted in reliance on an applicable statutory text-and-data-mining exception, fair use or fair dealing; and (b) the provider has not been enjoined by a court of competent jurisdiction from distributing the model or licensing its output for commercial use.
3. A documented safety-guardrail and red-teaming programme appropriate for the model's modality (audio, image, text), with published results at least annually.
4. An accessible opt-out / takedown mechanism for rightsholders who do not wish their works to appear in future training versions of the model, with SynthVerse as an escalation path where the provider does not respond within 30 days.

No pass-through liability. Except as expressly set out in the Terms of Service §10 and the provider's license agreement with SynthVerse, SynthVerse does not pass through to creators any liability of the base-model provider for infringement, defamation or third-party harm arising from a model's training set, and similarly does not waive any limitation of liability in the Terms of Service merely because a third-party model was involved in a dispute. Where a provider's own indemnity covers the creator for a specific infringement claim arising from model training provenance, SynthVerse will, on written request from the creator and with the creator's reasonable cooperation, facilitate the creator's access to that indemnity against documented third-party claims.

§8. Disputes Involving AI-Generated Content

Disputes relating to ownership, authorship, provenance, infringement or grade assignment of AI-assisted or fully-automated content are resolved under the following three-track process, which supplements (and does not replace) the general DMCA and dispute-resolution procedures described in the Copyright Policy and Terms §11.

Track 1 · Grade Re-Evaluation Appeal (internal)

If a creator disagrees with a Provenance Grade assigned by Vault's automated classifier or a manual AI Policy team adjustment, they may file an appeal within 14 calendar days of the grade decision by writing to ai-policy@synthverse.co.za with subject line "Grade Appeal: [release slug]". The appeal must include: (a) the release slug or Vault URN; (b) the creator's requested grade (A / B / C); (c) evidence of Human Creative Direction such as a Studio timeline project export, screenshot of rejected Forge candidates, FL Studio / Ableton / Logic project file, written lyrical draft progression, or a signed artist-declaration listing the creative choices exercised; and (d) any other supporting evidence. An appeals panel of three members (independent of the AI Policy team that issued the original grade) will issue a written, final decision within 10 business days. There is one appeal per release.

Track 2 · Rightsholder Training-Set Opt-Out / Takedown Request (provider)

If a rightsholder (copyright owner, performer, label, publisher) believes that a specific work of theirs was included in the training set of a Studio or Forge base model without license or applicable exception, and the relevant model provider has a documented opt-out mechanism listed in the Forge Model Register, SynthVerse will:

1. Direct the rightsholder promptly to that provider's opt-out web form or email contact, and — at the rightsholder's request — copy the provider's legal team on the referral;
2. Where a provider has confirmed to SynthVerse in writing that the specified work will be excluded from all future training runs and fine-tunes of that model, SynthVerse will record that exclusion in the Forge Model Register entry for the model;
3. Where applicable law allows, SynthVerse will — upon receipt of a completed DMCA-style notice targeting a specific published release and asserting that the release reproduces or is derivative of the complaining rightsholder's work — administer the standard notice-and-counter-notice process described in the Copyright & DMCA Policy, treating AI-generated releases identically to all-human releases for the purpose of that process.

Track 3 · Ownership / Authorship / Infringement Formal Dispute (external)

Formal legal disputes over ownership, authorship, copyright infringement or moral rights involving AI-generated content follow the general dispute-resolution ladder described in Terms §11: (1) informal good-faith resolution via disputes@synthverse.co.za within 14 + 30 days; (2) binding AFSA private mediation; (3) litigation before the Magistrate's or High Court of South Africa, Johannesburg, as the exclusive forum. Nothing in this AI Policy is intended to curtail a court's authority to make declaratory orders as to authorship, ownership or infringement status, and SynthVerse will adjust Vault grade badges, ownership records and DSP metadata promptly on receipt of a final, non-appealable court order, without requiring a separate dispute-resolution step.

§9. No Warranty and Liability for AI Output

Studio and Forge AI tools are provided on the same "AS IS" and "AS AVAILABLE" terms described in Terms §9 (Disclaimers). Without limiting those general disclaimers, the following specific disclaimers apply to AI output:

• No warranty of non-infringement for AI output. SynthVerse cannot and does not warrant that a given AI-generated output will be non-infringing with respect to the copyright, trademark, rights of publicity, or other intellectual property rights of any third party, in any jurisdiction. Base-model providers' contractual representations to SynthVerse about training-data provenance are passed through as set out in §7, but SynthVerse does not independently double-audit the entire training set of every base model (which can comprise millions of works) for copyright compliance, as that is not commercially feasible at current technology.
• No warranty of factual accuracy. Any factual or informational content generated by AI (for example: AI-suggested "liner notes" that include dates, biographical facts, historical references or musicological claims) is generated by probabilistic models and should not be relied on as factually correct without independent human verification. You are solely responsible for the factual accuracy of any material you publish on or through SynthVerse, whether AI-generated or fully human-written.
• No warranty of originality or copyright eligibility. SynthVerse does not represent that any given Grade C or D output is original or qualifies for copyright protection in any particular jurisdiction. Grade eligibility within SynthVerse (§3) is an internal platform-labelling convention only, not a legal opinion on copyright eligibility from SynthVerse or its legal advisors. If you require certainty about copyright eligibility for a specific work, you should obtain independent legal advice in your jurisdiction before relying on it commercially.
• Creator indemnification. Without limiting Terms §10, by publishing any release that uses Studio or Forge AI at any Grade, you agree to indemnify, defend and hold harmless SynthVerse, its affiliates, directors, officers, employees and contractors, and each base-model provider as a third-party beneficiary of this clause, from and against any and all third-party claims, damages, liabilities, costs and reasonable attorneys' fees arising out of or related to: (i) your breach of this AI Policy; (ii) the output you publish using Forge or Studio; (iii) any allegation that a Forge or Studio output you published infringes, misappropriates or violates any third-party intellectual property, publicity, privacy or other right; and (iv) any claim that your Grade self-assessment was knowingly materially false or misleading (e.g., intentionally selecting Grade A for clearly Grade D FAO to access Featured placement).

The liability cap in Terms §10.2 applies with full force to AI-Policy-related claims. To avoid doubt: claims relating to base-model training-data provenance that are brought against SynthVerse by third parties, and that SynthVerse is not able to pass through to the relevant provider under that provider's indemnity, count toward the same aggregate cap as all other claims under the Terms.

§10. Policy Evolution and Versioning

Both the technology of generative AI and the laws, regulations and industry practices governing it evolve rapidly. SynthVerse accordingly commits to revisiting this AI Ownership & Attribution Policy at least annually, and to issuing interim amendments whenever one of the following triggering events occurs:

1. A binding precedent from the Constitutional Court of South Africa, the Supreme Court of Appeal, the High Court of South Africa, the United States Court of Appeals for the Federal Circuit, the Court of Justice of the European Union, or a comparable apex or IP-specialist court in a jurisdiction where SynthVerse has > 5% of its active monthly creators, materially alters the legal treatment of authorship, ownership, training-copy liability or fair use/fair dealing of AI-generated works.
2. A new base model provider is onboarded to Studio or Forge whose output license or training provenance materially differs from the assumptions underlying this AI Policy (e.g., a provider that requires downstream attribution of base-model authorship on releases, or that reserves a per-stream royalty on output).
3. A new law or regulation is enacted — including but not limited to the South African pending Artificial Intelligence Governance Act, the European Union AI Act (Regulation 2024/1689), the United Kingdom AI (Intellectual Property) Act, or sector-specific amendments to the Copyright Act, 1978 — that imposes mandatory labelling, watermarking, disclosure, consent or training-data requirements on platforms offering generative AI tools to creators.
4. A security, safety or integrity incident involving AI on SynthVerse (e.g., a widespread voice-cloning abuse pattern, a training-data leakage event, or a Grade-inflation scam by a ring of accounts) that the AI Policy team determines requires a rule change to adequately protect creators and listeners going forward.

All amendments to this AI Policy are versioned by date. Material amendments (any amendment that reduces creator rights, narrows Grade A/B eligibility, increases indemnification obligations, or materially expands the scope of permitted training use of creator content) will be notified at least 30 calendar days in advance via: (a) email to every active creator account; (b) an in-app banner displayed on every login for 30 days; and (c) a prominent note at the top of this page linking to a diff-style comparison between the old and new versions. Non-material clarifications and typographical fixes may be applied without notice, with the "Effective date" at the top of the page updated accordingly.

The six most recent versions of this AI Policy are archived in the Legal Document Archive and are available on request to legal@synthverse.co.za.