VIEW THIS AS

Auto mode follows the Route Engine until you choose a viewpoint.

YOU ARE HERE

ROUTE CHECK

CONNECTED TO

WHAT NEXT

Use the canonical route for this room, or HELP if you are unsure.

Super Intelligence | Creative Industries, Authorship and Copyright in an Advanced-AI Economy

Three secondary students review open books and notes together in a classroom.

Creative industries, authorship and copyright as AI becomes more capable.

Separate contribution, ownership and scoped permission through complete fictional creative-project cases.

Three girls studying together with open books at a classroom table
Learning to connect creative contribution, evidence and responsible release decisions.

A creative work can be impressive, accurately credited and still lack a clear basis for its intended release. It can also contain a valuable human contribution even when part of its production involved AI. The useful questions are therefore more specific than “Who owns AI art?” Identify the material, the people and tools that contributed, the rights or permissions being claimed, and the exact use being proposed. An answer about one layer does not automatically settle the others.

This guide examines creative industries, authorship and copyright as AI systems become more capable. Super Intelligence is the series’ editorial umbrella; artificial superintelligence, or ASI, remains a hypothetical broadly superhuman capability. The future scenarios below are conditional. They do not claim that current systems are proven ASI, that creative work will disappear, or that a more capable model automatically becomes a legal author or rights holder.

The legal orientation is bounded to named sources and jurisdictions, checked on 1 October 2026. United States copyrightability guidance and Singapore ownership guidance answer different questions under their respective systems. They are not one global rule. The fictional project packets use invented permissions and an explicit internal release policy. Their release and hold answers follow those supplied terms, not a legal determination about a real work. For an actual consequential dispute or release, the relevant contracts, facts and current local law require qualified review.

Choose your reading route

Understand authorship and permissions · Read the Lantern project packet · Check amendments and changed uses · Try the independent Sable project

Concepts, jurisdictions and future possibilities

The complete Lantern packet and decisions

Changed uses, contribution and practice

Independent transfer and judgement

1. Separate creative contribution from permission to release

Imagine a short exhibition video combining a writer’s narration, an illustrator’s image, a generated texture, a musical composition and a recording of that composition. Calling the file “our video” may be convenient, but it conceals several different contributions and permissions. The producer might control the final edit without owning the music. The illustrator might have authorised a website appearance while withholding advertising use. A generated texture might be usable under a provider agreement without carrying an assurance of exclusive copyright ownership.

Creative contribution asks what each participant actually did. One person may have written the script, another selected and arranged images, and another performed music. Copyrightability asks whether particular material qualifies for protection under a relevant legal system. Ownership asks who holds the relevant rights. Permission asks whether the proposed use is authorised or otherwise legally permitted. Attribution asks how contributions and sources should be identified. Those questions interact, but none is a universal substitute for the others.

This separation becomes more important when AI makes variations inexpensive. A producer can create a hundred versions before noticing that the permission attached to one source image covers only a narrow use. The speed of generation does not expand that permission. Conversely, uncertainty about the protection of a generated component does not erase the need to recognise a writer’s original script or an editor’s documented contribution. Treat the assembled work as a set of relationships rather than one undifferentiated output.

A practical release decision should name the intended act. Printing 450 programmes, publishing a video on an owned website, placing a paid advertisement and uploading an illustration to a model are four different acts. A permission that covers one may not cover another. The audience, territory, duration, quantity and transformation can all matter under the actual terms. An asset is not permanently “cleared” in the abstract. It is assessed against a particular use under a particular evidence set.

The receiver of the analysis needs a decision they can inspect. “Looks safe” is not enough. A useful result says which version can proceed under which supplied permissions, which conditions remain attached, and what prevents other versions from proceeding. It should also distinguish a contractual checklist from an unresolved legal question. A producer can follow every printed teaching term while still learning nothing about a real court’s assessment of a different work. That boundary keeps the exercise useful and honest.

Back to contents · Next: 2. Read jurisdiction and date before reading conclusions

2. Read jurisdiction and date before reading conclusions

Copyright questions require a jurisdictional frame. The place of creation, the place of use, the parties’ agreements and the countries where protection or remedies are sought can raise different issues. This article does not resolve conflicts of law. Its narrower discipline is to label each legal source with the system it describes and avoid moving a conclusion into another system without support. A worldwide website does not turn all national laws into one uniform rule.

WIPO’s copyright guidance explains that national requirements matter for international protection and distinguishes licensing from simply making a work available. It also warns that rights analysis should consider limitations and exceptions before assuming that every unauthorised reproduction is infringement. The relevant lesson here is scope: identify the country and legal basis rather than treating a general copyright slogan as a complete answer. WIPO copyright questions

Dates matter in two ways. A legal source has a publication or update date, and a permission can have an effective period. A statement checked today may need updating after a legal change. A licence valid today may not cover next year’s campaign. Keep those clocks separate. The fact that a government webpage was recently refreshed does not extend a private agreement, and the fact that a contract remains in force does not establish that a legal interpretation has remained unchanged.

The United States material below concerns the Copyright Office’s analysis of AI-generated outputs and human authorship. The Singapore material concerns general ownership and commercialisation principles described by IPOS. Neither source is used to decide the fictional packets. Those packets stipulate authority, permissions and policy conditions so the reader can practise tracing a release decision without pretending to be a court. Their dates are invented operating facts, not predictions about future legislation.

A source note should therefore say more than “copyright law says.” Name the authority, jurisdiction, document and proposition being supported. If the question is about a contract, identify the actual version and clause instead. If the evidence does not settle the matter, state the unresolved question. This habit is especially valuable in fast-moving AI discussions, where a policy proposal, administrative guidance, court decision, vendor term and creator preference can otherwise be blended into one misleading claim about what is allowed.

Back to contents · Next: 3. United States guidance: human authorship remains a separate question

3. United States guidance: human authorship remains a separate question

The U.S. Copyright Office’s January 2025 report on copyrightability says that copyright protects human-authored expression, including qualifying human contributions in works that also contain AI-generated material. It does not extend protection to purely AI-generated material or insufficiently human-controlled expression. The report treats sufficiency case by case and says prompts alone, given the technology it examined, do not supply sufficient control. Creative human arrangements, modifications or perceptible expressive inputs can matter. These are United States conclusions, not global rules. Copyright Office Part 2 report

The practical question is not how many times someone pressed Generate. It is what human-authored expression is present and what evidence describes that contribution. A production record might preserve an original sketch, a manually written passage, a sequence of substantive edits or an arrangement decision. Such records can help describe the process. They do not guarantee a legal result. Quantity of effort, enthusiasm and tool use should not be substituted for the actual question the relevant authority is asking.

Nor should a project declare that every element becomes protected merely because a human assembled the file. Protection of an eligible human contribution and protection of each included component are different questions. A producer may need to distinguish the script, the arrangement and a generated background when making an ownership claim. A precise description can be less sweeping and more useful than a blanket claim over everything visible in the final export.

For the same reason, “AI was involved” is not a complete reason to deny every human contribution. An editor might use a tool for spelling suggestions while retaining authorship of the text. Another workflow might rely much more heavily on generated expression. The factual contribution record matters. This guide deliberately avoids a numerical threshold such as a percentage of words changed, because the cited report does not supply a universal percentage rule that would decide the cases for us.

Keep copyrightability separate from permission to use someone else’s inputs. A conclusion about protection of the output does not itself authorise copying an input work. Likewise, a licence to use an input does not promise that the resulting output will qualify for protection. These are separate links in the reasoning chain. The fictional exercise will record both the production history and the granted uses precisely, while leaving legal authorship conclusions outside its mechanical release test.

Back to contents · Next: 4. Singapore guidance: ownership and a licence do different jobs

4. Singapore guidance: ownership and a licence do different jobs

IPOS explains that, in general, the creator owns copyright, including commissioned work subject to contract; employment and other rules can affect the position. Its commercialisation guidance distinguishes assignment, which transfers ownership, from licensing specific uses. It recommends written terms identifying the parties, material, scope and obligations. The article uses those Singapore principles to explain why paying a creator and receiving a file should not be treated as identical to obtaining every right. IPOS ownership and commercialisation

IPOS also explains that copyright protects expression and that the copyright symbol is a notice rather than a source of additional rights. The absence of the symbol is not a waiver. That matters when a generated or downloaded asset arrives without a visible notice: appearance alone does not settle its rights position. This is a limited Singapore orientation, not an assertion that all works, contracts or countries follow identical rules. IPOS introduction to copyright

The production implication is to ask what was actually agreed. A commission might include one edition, specified channels or a defined territory. Another agreement might transfer specified rights. A fee invoice by itself may not describe the whole arrangement. Read the applicable agreement and any amendments rather than inventing terms from the size of the payment. In the teaching packet, the relevant grant is printed in full for the exercise so that no reader has to infer it from an invoice.

A licence can be entirely sufficient for the producer’s intended job without becoming ownership. If the goal is to include an image in one campaign, a suitable scoped permission may accomplish that goal. The creator can retain rights while the producer receives a dependable use. Conversely, a licence can be insufficient for a later merchandise line or model-input use. The failure is not that licensing is weak; it is that the new use differs from the permitted one.

This also changes how creative work should be handed over. The delivery should identify which rights are claimed, which uses are permitted and which obligations travel with the asset. A folder of polished files without the relevant terms can be less useful than a smaller package whose permissions are clear. As AI expands the number of variants, preserving those relationships becomes part of production quality. The rights record needs to follow the accepted asset, not remain attached only to an abandoned draft.

Back to contents · Next: 5. Attribution, provenance and ownership are not interchangeable

5. Attribution, provenance and ownership are not interchangeable

Attribution identifies a creator, contributor or source. Provenance describes where material came from and how it changed. Ownership concerns rights under the applicable law and arrangements. A project can have excellent attribution but inadequate permission; it can have permission but poor records of how an asset was transformed. A credit line is useful evidence of acknowledgement, but it does not manufacture a missing grant. Keeping these distinctions visible prevents a familiar ethical gesture from being mistaken for a complete legal basis.

Creative Commons provides a concrete example of permission with conditions. CC BY 4.0 permits sharing and adaptation, including commercial use, subject to its terms, including appropriate credit, licence information and change indication. Its legal framework does not promise every other permission a user might need, and the summary points to possible privacy, publicity or moral-rights limits. The actual legal code governs, not a casual “CC” label. CC BY 4.0 legal code

The fictional licences later in this guide are not Creative Commons licences. They have invented restrictions such as print caps and channel exclusions for teaching purposes. Do not infer that those restrictions belong to CC BY, or that adding an attribution line transforms a custom restrictive agreement into an open licence. The source example establishes a distinction; the packet supplies its own rules. Mixing the two would produce a false release decision.

Provenance should preserve asset identity. If an illustrator supplies image I1 version one and later supplies a revised image I1 version two, the terms may apply to one, both or neither without clarification. A checksum can help distinguish files, but it cannot prove that the person supplying a file owns its rights. A timestamp can document receipt, but it does not grant a licence. Technical records support identification and traceability; the permission still needs its own evidence.

A contribution description should be similarly careful. “Written by Nara; illustration by Leto; generated texture arranged by the editorial team” describes roles in our fictional project. It does not declare a legal determination about every layer. A human may deserve creative credit even when ownership lies elsewhere, and a producer may hold rights without personally creating every component. The public-facing credit and the internal rights ledger serve different readers. Both should be accurate, and neither should claim more than the evidence supports.

Back to contents · Next: 6. Why advanced AI could change creative bargaining

6. Why advanced AI could change creative bargaining

Consider a conditional future in which systems can generate convincing variations across text, image, sound and video much more quickly than current tools. The immediate change would be the cost and speed of producing candidates. The economic effect would still depend on what audiences value, which inputs are scarce, who controls distribution and how rights are allocated. Faster generation alone does not establish that every creator loses income, every producer gains, or every output becomes interchangeable.

One possible arrangement places value in trusted human direction. A publisher might pay for an identifiable writer’s judgement, a performer’s relationship with an audience or an artist’s distinctive body of work. Another might emphasise low-cost interchangeable content. These are alternative scenarios, not measured forecasts. The relevant question is which function the customer is buying: a usable image, an authentic performance, access to a catalogue, a particular person’s expression or a dependable cleared production package.

Rights can influence bargaining because permission to use a valuable catalogue or performance may remain scarce even when generating a substitute is technically easy. Whether a substitute is lawful, acceptable to the audience or commercially useful is a separate question. An agreement can also reserve future uses such as training, adaptation or endorsement. A more capable system may make those uses technically feasible, but the contract does not expand merely because the technology can now perform them.

Verification and provenance could therefore become more important in some markets. A buyer might value a package whose source records, contribution history and permitted uses are clear. That does not imply that every buyer will pay a premium or that provenance always proves authenticity. It identifies a possible service: reducing uncertainty about what the buyer can responsibly release. The service’s value depends on the buyer’s actual risks and alternatives, not on an assumed universal appetite for more documentation.

The analysis should also preserve the creator’s perspective. A broad assignment, a narrow licence and a recurring permission arrangement can distribute control and future opportunities differently. The right choice is not determined by AI capability alone. It involves the parties’ goals, negotiating position, project needs and relevant law. This guide does not recommend a contract for a particular reader. It teaches how to notice the decision variables before a technological story quietly decides them on everyone’s behalf.

Back to contents · Next: 7. The complete Lantern Studio project packet

7. The complete Lantern Studio project packet

Lantern Studio is an invented producer preparing a small exhibition campaign. Every person, asset, agreement, fee and date in this packet is fictional. The decision date is 20 October 2026. The internal policy requires a documented permission for every included asset and intended act, satisfaction of all printed conditions, a complete provenance record, an accurate credit block, and total committed project spending no greater than 1,000 credits. A credit is an invented unit, not a real currency or quoted market price.

For this exercise only, the named grantors have authority to grant exactly the printed permissions, the records are genuine and complete, and there are no unlisted third-party components or depicted people, brands or confidential materials in the accepted assets. These stipulations make the permission problem solvable. They are not conclusions a real producer may assume. The packet’s release test does not decide copyrightability, infringement, moral rights, privacy or any legal exception. An absent permission produces HOLD under the studio’s policy, not a judicial finding of illegality.

The initial expenditure is fixed: text permission and creation 120 credits; illustration commission and licence 350; composition licence 80; recording licence 60; generation service 20; editorial assembly 200. All six amounts are already committed and nonrefundable under the fictional budget rule. They total 830 credits. There are no taxes, transaction charges or other costs in the exercise. Proposed amendments create additional commitments only when their supplied signed versions become effective; no reader is asked to negotiate or purchase anything.

The base deliverables are P1, 450 printed programmes distributed only in Singapore on 20 October; W1, a 40-second video on the studio’s owned website, available worldwide from 22 October through 31 December 2026; and A1, a paid social advertisement displayed worldwide over the same period. P1 uses text T1 and illustration I1. W1 uses T1, I1, generated texture G1 and music excerpt M1 lasting 12 seconds. A1 uses T1, I1 and G1, with no music.

A fourth proposed act, U1, is uploading I1 to a generation service to create variations. It is evaluated separately from releasing P1, W1 or A1. The project has not performed U1. No real asset exists or is transmitted. All deliverables already satisfy the packet’s artistic, factual and technical checks unless a changed case states otherwise. That allows the reader to focus on permissions, provenance and budget without inventing hidden quality failures or treating those other checks as universally unnecessary.

Back to contents · Next: 8. The assets and their complete teaching permissions

8. The assets and their complete teaching permissions

Text T1 version one is an original fictional script written by Nara. Grant T grants Lantern Studio permission to reproduce, edit and include T1 in print, its owned website and paid social advertisements worldwide through 31 December 2026. There is no quantity cap. The required credit is “Text: Nara.” The grant does not transfer ownership and does not authorise model training or uploading T1 to a generation service. No later use is implied by payment of the 120-credit fee.

Illustration I1 version one is supplied by Leto for 350 credits. Grant I version one permits up to 500 cumulative printed copies in Singapore and display on Lantern Studio’s owned website worldwide through 31 December 2026. Proportional resizing is permitted; cropping and other alteration are not. Paid social advertisements and model-input uploads are expressly excluded. The required credit is “Illustration: Leto.” This is a scoped licence, not an assignment. The original source file and this exact grant version are retained in the packet.

Music M1 contains two separately documented layers. Grant C permits use of the composition by fictional composer Oren in excerpts no longer than 20 seconds within video on Lantern Studio’s owned website worldwide through 31 December 2026. Grant R permits use of the supplied recording, controlled by fictional record producer Vale, in excerpts no longer than 10 seconds in that same channel, territory and period. Both allow trimming an excerpt; neither allows remixing, paid advertising or other channels. Required credits are “Composition: Oren” and “Recording: Vale.” Both are necessary for this particular recording use under the exercise policy.

Texture G1 version one was generated from the studio’s original text description of abstract coloured shapes. No T1, I1, M1 or other third-party file was uploaded. The supplied fictional service terms permit commercial use of the resulting G1 file in the three listed channels worldwide without a time limit. They do not promise copyright protection, exclusivity or absence of every possible third-party claim. The studio’s policy accepts this bounded permission record for the exercise because no conflicting claim or unlisted input is present. Required production disclosure is “AI-generated texture; selected and arranged by Lantern Studio.”

Reference X1 is an unrelated downloaded illustration kept only as a hypothetical rejected candidate. Its creator, licence and permission status are unknown. It is not part of P1, W1 or A1, and no exception analysis is supplied. The policy prohibits including X1 in a release until the missing basis is resolved. A visible credit saying “found online” would not satisfy that requirement. This record exists to test whether a candidate can accidentally re-enter a later version after the accepted assets have already been checked.

Back to contents · Next: 9. Build the provenance record before deciding release

9. Build the provenance record before deciding release

The provenance record for T1 names Nara, version one, the retained original script and Grant T. It records that the studio shortened one sentence for W1 and that editing is permitted. The I1 record names Leto, version one and Grant I version one. It records proportional resizing for P1 and W1, with no cropping. The M1 record identifies both composition and recording, Grants C and R, and the chosen excerpt length. The G1 record identifies its original text-only input, service-terms snapshot and the studio’s selection and arrangement role.

These records answer different questions from the credits. The credit “Illustration: Leto” does not tell a reviewer whether 500 or 1,000 copies are permitted. The internal record does. The statement that a texture was AI-generated does not say whether a third-party image was uploaded to make it. The input record does. A complete handover should let the reviewer trace the accepted asset and its permitted use without reconstructing the entire creative conversation.

Versions should be attached to deliverables as well. P1 version one contains exactly T1 and I1. W1 version one contains exactly T1, I1, G1 and a 12-second M1 excerpt. A1 version one contains exactly T1, I1 and G1. If a designer later inserts X1 into W1, that is a new candidate version whose evidence differs. The old approval does not cover it simply because the filename still begins with W1. The asset manifest must describe the bytes or components actually being proposed for release.

For this paper exercise, exact version identifiers are sufficient; no actual file hash is required. In a real project, hashes or controlled asset IDs can help bind the manifest to files, while visual or audio inspection confirms that the manifest describes the content. Neither method replaces checking permissions. A correct hash of an unauthorised image proves which unauthorised image was used. Technical integrity and rights evidence are complementary controls, not interchangeable assurances.

The final record should preserve uncertainty explicitly. X1 has unknown source authority and therefore cannot become “original studio material” merely because a tool transformed it. G1 has a documented permitted-use basis but no guaranteed exclusive copyright claim. Those are different states. The project can accept a nonexclusive component under its stated policy while rejecting a component with no usable permission record. The decision depends on the actual release requirements rather than a blanket preference for either human-made or generated assets.

Back to contents · Next: 10. Decide the original release package

10. Decide the original release package

P1 passes the packet’s permission test. Its 450 copies are below Grant I’s cumulative cap of 500, distribution is confined to Singapore, and the date falls before the end of 2026. T1 permits print use worldwide, which includes the specified Singapore distribution. I1 is only proportionally resized, so the permitted transformation is respected. The P1 credit block contains Text: Nara and Illustration: Leto. With the supplied provenance and other checks complete, the policy result is RELEASE P1 version one.

W1 does not pass in its original form. T1, I1 and G1 each have a basis for the owned-website use. The composition grant allows a 12-second excerpt because 12 is no more than 20. The recording grant permits only 10 seconds, so the same 12-second excerpt exceeds that layer’s limit by two seconds. The credits can all be correct and the video can still fail. The result is HOLD W1 version one for recording-duration mismatch, not a claim that the composition permission is missing.

A1 also fails, for a different reason. T1 permits paid social advertising and G1’s supplied service terms permit the listed commercial channel. Grant I version one expressly excludes paid social advertisements. Worldwide owned-website permission does not include a different platform and advertising purpose merely because both are online. The result is HOLD A1 version one for I1’s excluded channel. Adding a larger illustration credit does not repair the missing permission.

U1 is not permitted under Grant I version one because model-input uploading is expressly excluded. The project does not perform it. This remains true even if the intended generated variations would later be used only on the owned website. Permission for a final display does not erase a separate restriction on an intermediate act. The policy result is HOLD U1. The exercise is about identifying that distinction, not establishing the legality of model training or uploads in every jurisdiction.

At the package level, one releasable deliverable does not make the whole campaign cleared. The report should show P1 released under the packet, W1 held for the recording limit, A1 held for the illustration channel, and U1 held for the explicit upload exclusion. Total committed spending remains 830 credits, leaving 170. Budget availability does not repair a rights mismatch by itself. It only indicates whether a separately supplied amendment or redesign could fit the studio’s additional internal constraint.

Back to contents · Next: 11. Repair the exact mismatch without inventing permission

11. Repair the exact mismatch without inventing permission

The simplest supplied repair for W1 is editorial: shorten M1’s excerpt from 12 seconds to 10 seconds. The packet explicitly states that this change preserves the video’s artistic and technical acceptance and costs nothing extra. Both music grants allow trimming. At 10 seconds, the composition and recording limits are satisfied. W1 version two contains the same other assets and complete credits, so it passes the permission test within the original 830-credit budget.

This repair works because it changes the use to fit an existing grant. It does not reinterpret 12 seconds as “approximately ten,” ignore a protected layer, or assume that a small overage is harmless. The exact fictional terms make the answer deterministic. Real agreements may contain definitions or exceptions that require interpretation, but this packet supplies none. The reader should solve the problem that exists rather than importing an imagined tolerance because it would make the preferred edit easier.

The A1 problem cannot be repaired by shortening music, because A1 has no music. It cannot be repaired by changing the image size, because proportional resizing was already permitted. Its failing variable is the advertising channel. A valid repair would require a different authorised asset or a supplied grant that includes that channel. The next chapter provides such an amendment. Until then, the accurate answer is a hold with a precise reason, not a promise that the rights owner would probably agree.

The U1 exclusion also requires its own analysis. A grant permitting advertising would not automatically permit uploading to a model. These uses can be negotiated separately in a fictional or real agreement. The studio should not convert an unrelated amendment into broad permission for every technically available transformation. This is one of the main lessons for an advanced-AI economy: expanding what systems can do makes precise use definitions more important, because new acts may fall outside older language.

Repair records should retain the old failure. W1 version one remains held; version two is the corrected candidate. That history prevents the longer excerpt from reappearing in a later export under an old approval label. It also explains why the accepted version differs from the original artistic draft. The record need not preserve every rejected experiment forever, but it should retain enough evidence to distinguish the permitted release from the candidate that failed. Creative speed is useful when version discipline keeps up with it.

Back to contents · Next: 12. Changed contracts create new, bounded conclusions

12. Changed contracts create new, bounded conclusions

The packet now supplies a signed Amendment I2, effective 21 October 2026, with an additional nonrefundable fee of 150 credits. For I1 version one, it raises the cumulative print cap from 500 to 1,000, permits worldwide print distribution and paid social advertisements, and permits cropping as well as proportional resizing. It extends those licensed uses and the owned-website use through 31 March 2027. The credit requirement remains. Model-input uploads remain expressly excluded. All other printed conditions continue unchanged.

The exercise stipulates that Amendment I2 is validly executed by the authorised fictional parties and supersedes only the conditions it expressly changes. We are not deciding whether a real signature or contract is legally effective. That stipulation matters because an unsigned proposal would not be equivalent. The decision date for amended releases is now 22 October 2026, after its effective date. Do not retroactively describe a proposed paid advertisement on 20 October as covered by the later amendment.

With I2 effective, A1’s illustration channel mismatch is repaired. T1 and G1 already permit that channel, and the credits and provenance are complete. A1 version one can therefore pass the packet’s permission test for its specified October-to-December period. U1 remains held because the upload exclusion was expressly preserved. This contrast tests clause-level reading: one new document can change several permissions while leaving another restriction untouched.

An optional signed Amendment R2 to the recording grant is also supplied, effective 21 October, for 40 additional credits. It raises the allowed M1 recording excerpt on the owned website from 10 to 15 seconds through 31 December 2026. It changes no channel, territory, credit or remix condition. With R2 and the original composition grant, a 12-second excerpt would satisfy both music layers. The amendment does not allow paid advertising or an excerpt longer than the unchanged composition limit. The amendment comparisons below are independent scenario branches: only the amendments named in a branch are executed, signed and effective in that branch. In the I2-only branch, R2 is never executed and total spending is 980 credits; in the I2-plus-R2 branch, both are executed and total spending is 1,020 credits. No branch cancels or refunds an existing commitment, and an already signed, effective R2 would remain a nonrefundable 40-credit commitment.

Contractual fit and internal budget must still be checked separately. I2 alone brings spending to 980 credits. Adding R2 brings it to 1,020. Under the fixed 1,000-credit cap, the two-amendment package is held for budget even though the specified uses could fit those grants. The studio can instead adopt I2 and W1 version two’s 10-second excerpt for 980. This is an internal release-policy decision, not a statement that exceeding an internal budget determines copyright infringement.

Back to contents · Next: 13. Reproduce the quantities and costs

13. Reproduce the quantities and costs

The initial expenditure is 120 + 350 + 80 + 60 + 20 + 200 = 830 credits. The remaining allowance is 1,000 − 830 = 170. I2 costs 150, leaving 20 after its addition. R2 costs 40. Therefore 830 + 150 + 40 = 1,020, which exceeds the cap by 20. The alternative using I2 and the no-cost 10-second edit totals 980, which remains 20 below the cap. No refund from the original grants is available under the packet.

Now introduce a separate reprint request for 80 more P1 programmes after the original 450 have been printed. The licence cap is cumulative, not per print run. The total becomes 450 + 80 = 530. Under Grant I version one, that exceeds 500 by 30 copies. Printing 80 is not permitted merely because 80 is smaller than 500. The relevant unit is total copies across the covered uses of I1 in this project, as the packet explicitly defines it.

With I2 effective, the same 530 total is within 1,000, leaving 470 copies of contractual capacity. That headroom is not an instruction to print 470 more. It is the unused part of a limit. A new print decision would still need its own project purpose, date, distribution and approval checks. The distinction between permission and demand matters economically: paying for a larger cap can create an option without making it useful to exercise every part of it.

For the music, the base maximum is the smaller of 20 seconds for composition and 10 for recording, so the usable excerpt limit is 10 seconds for this combined asset and channel. With R2, it becomes the smaller of 20 and 15, or 15. A proposed 18-second excerpt would still fail the amended recording limit by three seconds, even though it fits the composition grant. The whole asset must satisfy every required layer; averaging the two limits would have no basis.

These calculations are deliberately modest. They let the reader inspect how a rights condition becomes a release constraint without pretending that legal judgement is arithmetic. Numbers settle the supplied caps and budget. They do not settle human authorship, legal originality, fair use, validity of a real grant or the truth of an ownership claim. Keep the calculation attached to the specific contractual or policy variable it measures. Exact arithmetic becomes misleading when its meaning is allowed to drift.

Back to contents · Next: 14. A change in date, territory or input can reopen the decision

14. A change in date, territory or input can reopen the decision

Move A1’s display period to February 2027 while keeping the same assets. I2 now covers I1 through March, and G1 has no time limit in the supplied terms. Grant T still ends on 31 December 2026. The changed A1 therefore fails on T1’s term. One extended asset does not extend the whole package. A release ledger needs to track the intersection of the included permissions, not merely the latest expiry date found anywhere in the folder.

Move the original P1 distribution to Canada on 20 October, before I2 takes effect. T1 permits worldwide print use, but Grant I version one limits print distribution to Singapore. That candidate is held for territory. Once the valid amendment becomes effective, worldwide print permission changes that answer for later uses within the amended period. The task must preserve which date and grant version are being tested. Combining the original date with the later territory allowance creates an unsupported hybrid.

Change G1’s production history so that I1 was uploaded to generate the texture. The original clean-input stipulation no longer applies. Grant I excludes that upload both before and after I2. The studio’s policy must hold the affected candidate for unresolved input-permission compliance, even if G1’s service terms permit commercial output use. A provider’s output clause cannot retrospectively grant rights in an input it does not control. The packet’s authority assumption belongs to each named grantor and their specified material.

Insert X1 into a new W1 version as a background detail. The accepted video does not remain cleared merely because the new image is small or visible briefly. Under the supplied policy, X1’s permission basis is unknown and no exception analysis exists. The candidate is held pending resolution or removal. If X1 is removed and the accepted asset manifest restored, the earlier permission analysis can become relevant again after verifying the actual new version. A label saying “final” cannot substitute for that comparison.

These changes show why a rights decision should have a scope and a review trigger. New channel, new territory, new date, new transformation, new input and new component are all potentially material. Not every minor export change needs a completely new legal analysis, but a meaningful use change should be checked against the evidence it affects. A more capable AI could help identify those dependencies. It should not silently decide that the old approval is close enough because the new variant looks similar.

Back to contents · Next: 15. Describe human contribution without making unsupported ownership claims

15. Describe human contribution without making unsupported ownership claims

Lantern’s contribution record says Nara wrote T1, Leto created I1, Oren composed the music and Vale supplied the controlled recording. The studio selected and arranged G1, edited the script within its permission and assembled the final deliverables. Those are factual roles stipulated by the exercise. The record does not declare that every role creates a separate copyright, nor that one participant owns the entire assembled campaign. That legal question is outside the packet’s mechanical permission test.

For AI-assisted work, a useful record can distinguish initial material, generated suggestions, accepted components and human changes. Suppose an editor asks for alternative transitions but writes the final passage independently. That process differs from accepting an entire generated passage unchanged. Neither description should be inflated into a legal conclusion without the relevant analysis. The record’s purpose is to preserve evidence accurately so that later reviewers do not have to choose between a misleading “all human” label and an equally misleading “all AI” label.

Avoid using effort as an automatic ownership measure. A person can spend many hours selecting a result, while another can make a brief but significant expressive contribution. The legal relevance of those acts depends on the applicable framework and facts. The article’s United States source focuses on human expression and control, not a universal stopwatch or prompt count. A project can document effort for payment and management purposes while recognising that effort and copyrightability answer different questions.

The same caution applies to exclusivity. A service term may allow a user to exploit an output without promising that no other user will receive similar material. A client may care strongly about exclusive assets; another may accept shared or nonexclusive components. The production brief should state that requirement rather than discovering it after delivery. In Lantern’s packet, the policy permits G1 despite the lack of an exclusivity promise. Changing the brief to require exclusive rights in every component would reopen that decision.

Public descriptions should reflect the receiver’s actual need. An exhibition audience may need clear creative credits and a truthful note about generated material. An internal reviewer may need the detailed contribution and permission ledger. A legal adviser may need the complete agreements and production evidence. Giving every audience the same enormous file is not automatically better transparency. The essential requirement is that each claim can be supported and that important uncertainty is not hidden behind a polished credit line.

Back to contents · Next: 16. Training, transformation and distribution are different uses

16. Training, transformation and distribution are different uses

An AI workflow can involve several stages before a public output exists: gathering reference material, uploading it to a service, generating variations, editing them, storing them and distributing a final work. Permission for one stage does not automatically answer every other stage. The Lantern upload example isolates this point by giving I1 a clear display licence and a clear input-use exclusion. The answer follows the supplied contract, without claiming that all model inputs are treated the same way under actual law.

Training a model and using an existing model to transform one input are also not identical technical or legal descriptions. A producer should establish what the service actually does with supplied material and which terms apply, rather than labelling every interaction “training.” This article does not evaluate a real provider’s retention or training practices. G1’s fictional terms are intentionally limited to the permissions printed in the packet. No additional promise about confidentiality, deletion or data use should be inferred.

The Copyright Office’s AI initiative treats digital replicas, output copyrightability and generative-AI training in separate report parts. That organisation itself is a useful reminder that these are distinct policy questions. Its current initiative page identifies Part 3 as a pre-publication training report, rather than turning the output report into a universal answer about training. This article does not attempt to decide a training dispute or summarise all litigation. Copyright Office AI initiative

The producer’s immediate task is often narrower: determine whether the proposed workflow fits the permissions and obligations actually supplied. If an agreement prohibits uploading a commissioned image, a model’s ability to produce a beautiful adaptation is irrelevant to that contractual mismatch. If the material has an independently established permitted basis, the workflow still needs to preserve its conditions. The conclusion should name the basis rather than using the generation method as a blanket defence or accusation.

A future ASI system might identify more potential sources, transformations and distribution routes. That could increase useful options, but it could also multiply the number of rights relationships a project must understand. The appropriate response is not to assume omniscience. It is to preserve authority, provenance and review at the points where uses change. A system’s prediction that permission would be granted is not permission, and a proposed contract is not an executed one merely because its wording is plausible.

Back to contents · Next: 17. Creative abundance does not eliminate production responsibility

17. Creative abundance does not eliminate production responsibility

Under the hypothetical high-capability scenario, generating another draft could become almost frictionless while deciding what to release remains consequential. The production constraint may shift toward selection, audience trust, rights evidence or distribution. This is a conditional mechanism, not a measured claim about today’s entire creative economy. Different sectors can experience different effects. A custom illustration, a licensed music catalogue and a factual educational resource do not have identical customers or value propositions.

One possible producer strategy is to minimise dependence on uncertain components. Another is to pay for distinctive licensed work because it improves the project. A third is to combine human-authored material with generated components whose role is clearly disclosed. These choices should be evaluated against the actual brief and permissions. The cheapest generation is not necessarily the cheapest completed project if it creates substantial review, replacement or dispute costs. Equally, an elaborate rights process is wasteful if it records irrelevant detail while missing the actual release condition.

Creators may also choose different relationships with AI providers and producers. Some may license particular uses, some may reserve them, and some may develop new collaborative practices. The article does not predict which arrangement will dominate. It asks who can decide, what they are granting, what they retain and how changes are recognised. Those questions remain meaningful even when a system’s technical ability expands. Creative capability does not dissolve the parties’ different interests.

Audience trust is another distinct variable. A fully permitted synthetic texture can still be presented misleadingly if a caption claims it is a documentary photograph. A correctly credited human work can still be distributed outside its permission. Rights, authenticity and truthful presentation should therefore remain separate review questions. Passing one does not excuse failure on another. In Lantern’s packet, factual and technical acceptance are stipulated so the permission exercise is tractable; real production must establish those facts rather than assume them.

A useful industry-level analysis should therefore follow the complete accepted outcome. Who receives the work? What value do they seek? Which contributions are necessary? Which permissions support delivery? Who bears uncertainty and correction costs? How can the agreement be updated when a new use appears? Those questions make the future discussion concrete without turning it into a claim that one technology determines cultural value, legal ownership or the terms of every creative relationship.

Back to contents · Next: 18. Independent practice with the Lantern packet

18. Independent practice with the Lantern packet

Exercise one asks whether P1’s 450 Singapore copies can proceed under the original grants on 20 October 2026. Name every included asset and condition that matters. Then change only the quantity to 520 and give the new result. Do not change the territory, date or grant version. The purpose is to identify the failing variable precisely, rather than describing the entire project as generally risky.

Exercise two asks why W1’s original 12-second music use is held even though the composition grant allows 20 seconds. Give the allowed combined limit before and after R2, and explain why 18 seconds still fails after that amendment. Exercise three asks whether a complete credit block repairs A1’s original paid-advertising problem. Identify the missing permission and the smallest supplied contractual change that addresses it.

Exercise four gives I2 and asks whether U1 is now permitted. List the conditions that I2 changes and the relevant exclusion it preserves. Exercise five asks for the lowest-cost supplied package that releases P1, W1 and A1 within the 1,000-credit cap. Use only the supplied amendment and editing options; invent no discount, refund or substitute asset. Show the total and remaining allowance.

Exercise six adds the 80-copy reprint after 450 copies have already been made. Calculate the cumulative total, the original-cap excess and the amended-cap headroom. Exercise seven moves A1 to February 2027 with I2 effective. Identify which included asset’s term now prevents release. Exercise eight inserts X1 into the otherwise accepted W1. Explain what the existing approval can and cannot establish about the changed version.

Exercise nine changes G1’s input history to include an upload of I1. Decide the policy result before and after I2 and distinguish the input restriction from G1’s commercial-output permission. Exercise ten changes the client brief to require exclusive rights in every component. Identify why the original G1 acceptance needs reopening. Your answer should not declare G1 unlawful or unprotectable; it should name the missing assurance relevant to the changed requirement.

Back to contents · Next: 19. Lantern answers and the evidence that settles them

19. Lantern answers and the evidence that settles them

Answer one: P1 passes because T1 permits print use and I1 permits 450 cumulative copies in Singapore before year-end, with proportional resizing and the required credits. At 520 copies, the candidate exceeds I1’s original 500-copy cap by 20 and is held. The text permission is still sufficient. The diagnosis concerns the illustration quantity condition, not the presence of AI elsewhere in the campaign.

Answer two: both music layers must fit. The base combined limit is min(20,10) = 10 seconds. R2 changes it to min(20,15) = 15. Eighteen seconds exceeds the amended recording allowance by three. Answer three: attribution does not add the excluded paid-social channel. I2 is the supplied amendment that grants that use for I1, effective only from its stated date. Correct credits remain necessary after the channel is added.

Answer four: U1 remains held. I2 changes print quantity, print territory, paid-social use, cropping and the term, but expressly retains the model-input exclusion. Answer five: use I2 and W1 version two’s accepted 10-second edit, without R2. The total is 830 + 150 = 980, leaving 20. The package with both amendments costs 1,020 and fails the budget policy even though the 12-second website use would fit the amended music terms.

Answer six: 450 + 80 = 530. That exceeds the original cap by 30 and leaves 1,000 − 530 = 470 under I2. Answer seven: T1 still expires on 31 December 2026, so the February advertisement is held despite I1’s extended term and G1’s unlimited period. Answer eight: X1 creates a new component with no documented permission basis. The old manifest and approval do not cover it. Remove it and verify the restored candidate, or resolve the missing basis through an authorised process.

Answer nine: both before and after I2, the altered G1 candidate is held under the studio’s input-permission policy because I1’s upload exclusion remains. The service’s permission to use an output does not grant rights in Leto’s input. Answer ten: the fictional service terms do not promise exclusivity, so G1’s evidence does not satisfy the new brief. The correct result is a hold for the missing assurance or a redesigned brief/component, not an unsupported legal conclusion about every generated texture.

Back to contents · Next: 20. Independent transfer: the Sable audio-and-booklet project

20. Independent transfer: the Sable audio-and-booklet project

Sable Studio is a second fictional producer, evaluated on 15 November 2026. Its policy again requires documented permission for every included component and use, complete credits and provenance, and no unresolved condition. Its budget is 600 credits. All named grantors are stipulated to have authority over the specified material. All records are complete except the explicitly missing music-recording grant. There are no real people depicted, no confidential inputs and no unlisted assets. Legal validity and copyrightability are outside the exercise test.

Script S1 is written by Eda. Its 140-credit grant permits editing and use in Sable’s owned website and distributed audio podcast worldwide through 30 November 2026, with credit “Script: Eda.” It does not initially permit a downloadable PDF. Voice recording V1 is performed and supplied by fictional performer Fen under a 160-credit grant covering this recording and the performer’s supplied permission for excerpts up to six minutes in the distributed audio podcast worldwide through the same date. Credit is “Voice: Fen.” No voice cloning, model-input upload or other use is authorised.

Illustration B1 is created by Garo. Its 90-credit grant permits 200 printed programmes in Singapore through 30 November, with proportional resizing and credit “Illustration: Garo.” It does not permit downloadable PDFs. Music composition C2 has a 40-credit grant for excerpts up to 45 seconds in the podcast worldwide through 30 November, credited “Composition: Hali.” A particular recording R3 is proposed, but its rights controller and permission are unknown. No recording licence or legal-exception analysis is supplied. Editorial assembly costs 100 credits. These five committed costs total 530 and are nonrefundable.

Episode E1 contains S1, 312 seconds of V1 and 30 seconds of the proposed R3 recording of C2. All timing and technical checks pass, but the recording permission is missing. The supplied no-cost alternative E2 removes the music entirely and is artistically accepted; it contains S1 and the same 312 seconds of V1. Booklet F1 is a worldwide downloadable PDF containing S1 and B1. It currently has no permission for either component in that format. No copies or downloads have occurred except those explicitly introduced below.

Two signed amendments, effective 15 November, are supplied. Amendment S adds worldwide PDF use for S1 through 30 November for 20 credits. Amendment B adds worldwide PDF use for B1 through 30 November for 30 credits, capped at 500 cumulative downloads; the existing print permission remains separate and unchanged. Credits and proportional-resizing conditions continue. With both amendments, 120 F1 downloads have already occurred in the changed case, and a new distribution request would add 400. Decide E1, E2 and F1; calculate voice headroom and amendment spending; then evaluate the 400-download request and a proposed release on 1 December.

Back to contents · Next: 21. Transfer answers and changed assumptions

21. Transfer answers and changed assumptions

E1 is held because the proposed R3 recording lacks a documented controller and grant. C2’s composition permission covers 30 seconds within its 45-second limit, but it does not supply the missing recording layer. No amount of credit to Hali repairs that absence. The exercise does not determine that the recording use is legally infringing; it determines that the supplied policy lacks the evidence required to release this candidate.

E2 passes for release on 15 November under the supplied podcast permissions, assuming its stated credits and provenance are complete. Six minutes equals 360 seconds. The voice use is 312 seconds, or five minutes twelve seconds, leaving 48 seconds within V1’s excerpt cap. The absence of music in E2 removes the unresolved component rather than pretending its permission exists. The already committed 40-credit composition fee remains part of spending because the packet makes it nonrefundable.

F1 is initially held because both S1 and B1 lack PDF permission. Applying only one amendment leaves the other component unresolved. With both effective amendments, spending becomes 530 + 20 + 30 = 580 credits, leaving 20 under the 600-credit cap. F1 can be released within the amended conditions through 30 November, but the cumulative download limit still applies. The print allowance does not increase or reduce that separate PDF limit.

After 120 downloads, a further 400 would produce 520, exceeding the 500-download cap by 20. At most 380 additional downloads fit the supplied cap. That is a permission limit, not a recommendation to distribute exactly 380 or a claim about an actual platform’s ability to enforce it. The 400-download request is held unless its scope changes or another valid permission is supplied. A distribution system would need an appropriate operational control; the paper exercise does not pretend that a written limit enforces itself.

On 1 December, the relevant S1, V1 and B1 permissions in the packet have expired. The proposed new episode or PDF release cannot rely on the November terms. The amendments changed format permissions and quantity, not the end date. This is the independent transfer lesson: the structure resembles Lantern, but the failing components, units, channels and dates differ. A good reviewer reconstructs the decision from Sable’s own records instead of copying Lantern’s answer or assuming that every amendment removes every restriction.

Back to contents · Next: 22. Use the decision record to preserve choice

22. Use the decision record to preserve choice

A complete release record should identify the deliverable version, included assets, proposed acts, governing evidence set, unresolved questions and final decision. It should state whether the result is a contractual-policy pass, a legal opinion or a technical acceptance. In the fictional packets, only the first and stipulated technical acceptance are present. That labelling prevents a downstream collaborator from quoting RELEASE as if a lawyer had certified every possible use in every country.

A hold should be equally specific. “Missing R3 recording grant” is actionable. “Copyright problem” is too vague. “T1 term ends before the proposed February campaign” explains what changed. “AI content unsafe” misidentifies the issue. Specific holds preserve options: change the use, remove a component, obtain the appropriate evidence, or ask the relevant authorised person to decide. They also avoid unnecessary work on assets that already satisfy their conditions.

For creative workers, a clear record can preserve the scope of what they agreed to. For producers, it can support dependable delivery. For audiences, truthful credits and disclosures can explain the work they encounter. These benefits are related but not identical. A producer’s desire for efficiency does not decide an artist’s preferred rights arrangement, and an audience’s preference for a style does not establish ownership. The goal is an inspectable relationship in which the parties can understand and contest the claims being made.

If future systems become much more capable, the same record can help distinguish capability from authority. A system may identify a cheaper edit or a promising new channel, but it cannot treat a predicted agreement as an actual grant. It may organise contribution evidence, but it should not invent human authorship facts. It may detect a missing term, but it should not quietly fill it with a convenient assumption. Better assistance should make the decision clearer to the people who own it.

The strongest sign of understanding is independent transfer. You can identify the exact use, follow each required component, reproduce the relevant limits and costs, preserve the distinction between permission and legal protection, and explain what would change the decision. You can also say when the packet does not answer the legal question. That competence is more durable than memorising a slogan about AI and copyright, because new tools, contracts and jurisdictions can change while the discipline of scoped evidence remains useful.

Back to contents · Next: 23. Questions readers commonly ask

23. Questions readers commonly ask

Does paying for an AI service mean I own every output? The answer depends on the relevant terms, applicable law and facts. A contractual permission to use an output should not be restated as a guarantee of copyright protection or exclusivity unless the evidence supports that claim. G1’s fictional terms deliberately permit commercial use while withholding those guarantees. The exercise shows why a producer should preserve the distinction rather than infer a stronger promise from payment alone.

Is attribution enough? It answers a different question from permission. A required credit can be one condition of a valid licence, but writing a creator’s name does not expand an excluded channel, extend a term or supply an unknown recording grant. Conversely, having permission does not make every credit optional. Check both the use basis and the attribution conditions that actually apply. The Lantern advertisement and Sable recording examples fail despite possible accurate credits.

Can I call an AI-assisted work entirely human-made because I selected the result? Describe the actual process accurately. Selection, arrangement, drafting and generation are different contributions. Whether particular human activity supports a legal authorship claim requires the relevant jurisdiction and facts; a production label should not conceal material generation or invent expressive work that did not occur. A precise contribution description is more useful than a sweeping label designed to settle every question at once.

Are all uses without an individual licence forbidden? No. Actual legal systems can include exceptions and limitations, and some material may be in the public domain or subject to an applicable open licence. This article’s fictional policy intentionally requires the printed permission basis and supplies no exception analysis, so its HOLD labels are narrower than legal conclusions. Do not export the exercise’s conservative rule into a claim that no other lawful route exists in reality.

Will ASI make copyright irrelevant? That is a speculative claim, not a consequence established by stronger generation. Even under a hypothetical major capability increase, societies and parties would still face choices about incentives, control, access, attribution and responsibility. Laws and agreements could change, but the direction is not fixed by technical capability alone. The useful analysis identifies alternative arrangements and their consequences while keeping current legal facts, contractual assumptions and future possibilities visibly separate.

Back to contents · Next: Sources and scope

Sources and scope

Legal orientation checked on 1 October 2026: the U.S. Copyright Office’s January 2025 Part2 report and current AI initiative page; Singapore IPOS ownership/commercialisation and copyright-introduction guidance; WIPO copyright questions; and the CC BY4.0 legal code. Each is linked beside the limited proposition it supports. The fictional grants are not legal templates, Creative Commons licences or assessments of real rights. No real file, licence, identity or confidential material was transmitted for the exercises.

For the distinct question of novelty, taste and human expression, read Super Intelligence and Creativity. For the practical whole-project workflow, use How to Complete an Entire Creative Project With Super Intelligence. The existing Media Rights guide and Media Attribution guide explain the general reuse and source-lineage layers. Return to What Is Super Intelligence? for the wider series.

Back to contents · Return to the Super Intelligence master guide