Copyright & Intellectual Property: The Expert Guide

Copyright Fundamentals vs. Plagiarism: Legal Boundaries and Ethical Distinctions

Most professionals conflate copyright infringement with plagiarism, treating them as interchangeable concepts. They are not. One is a legal violation enforceable in federal court with statutory damages up to $150,000 per work. The other is an ethical breach that can end careers and academic records without a single lawsuit being filed. Understanding where these two frameworks diverge — and where they overlap — is foundational to any serious intellectual property strategy.

Copyright: A Bundle of Exclusive Rights, Not Just "Ownership"

Copyright protection attaches automatically the moment an original work is fixed in a tangible medium of expression. Under 17 U.S.C. § 102, this covers literary works, musical compositions, audiovisual content, software code, and architectural works, among others. Registration with the U.S. Copyright Office is not required for protection to exist, but it is required to file an infringement lawsuit and to claim statutory damages — a critical practical distinction most creators overlook. Registration within 90 days of publication preserves the full range of remedies.

What copyright actually grants is a bundle of exclusive rights: reproduction, distribution, public display, public performance, and the creation of derivative works. Infringing any one of these — even if you credit the original author — constitutes copyright infringement. This is the first major divergence from plagiarism: attribution is legally irrelevant to infringement. A documentary filmmaker who licenses footage correctly but exceeds the licensed distribution territory has infringed copyright. A student who copies three paragraphs verbatim but cites the source has committed plagiarism.

Plagiarism: Ethical Standards Operating Outside the Law

Plagiarism is governed by institutional policies, professional codes of conduct, and reputational norms — not statutes. A journalist who fabricates quotes or lifts sentences from a competitor without credit faces termination and industry blacklisting, not a federal injunction. Academic institutions typically define plagiarism to include subtler forms of intellectual misrepresentation beyond verbatim copying, such as paraphrasing without attribution, submitting AI-generated text as original work, or recycling one's own prior work without disclosure (self-plagiarism).

The critical asymmetry practitioners must internalize: public domain works can be copied verbatim without copyright liability, but presenting that copied text as your own original writing still constitutes plagiarism. Shakespeare's sonnets carry zero copyright risk — claiming you wrote them carries significant professional risk. Conversely, a ghostwriter who produces content contractually transferred to a client commits no plagiarism under standard work-for-hire agreements, even though the published author wrote nothing.

Practical implications for content strategies, legal compliance programs, and publishing workflows follow from keeping these frameworks distinct:

• Copyright clearance is a legal question requiring license review, fair use analysis, and registration verification
• Attribution practices address ethical standards set by editorial guidelines, academic integrity policies, or professional associations
• Work-for-hire and assignment agreements resolve ownership but do not eliminate plagiarism concerns in contexts where original authorship is publicly represented
• Fair use (17 U.S.C. § 107) may legally permit reproduction without permission, but the ethical obligation to credit sources remains independent of that legal defense

Operating a content business, legal department, or academic program without this distinction firmly established creates compounded exposure — legal liability on one track and reputational damage on the other, often simultaneously.

Intellectual Property Rights in Academic Research: Ownership, Attribution, and Disputes

Academic research generates some of the most contested intellectual property on the planet. A single biomedical study can be worth hundreds of millions in licensing revenue, yet the default legal framework governing who actually owns that work is riddled with ambiguity. The core tension lies between three competing claimants: the researcher who generated the ideas, the institution that provided funding and infrastructure, and the publisher who disseminated the findings.

Who Owns What: The Institutional IP Trap

Most researchers sign employment or enrollment agreements that contain IP assignment clauses transferring ownership of any work-related inventions and research outputs to the institution. In the United States, the Bayh-Dole Act of 1980 fundamentally shaped this landscape by allowing universities to retain patent rights over federally funded research — rights they previously couldn't hold. As a result, institutions like MIT and Stanford collectively earn over $100 million annually in licensing revenues. What many graduate students and early-career researchers fail to realize is that their thesis work, software tools, and even lab notebooks may fall squarely under institutional ownership from day one. Anyone entering an academic environment should review their IP agreement before producing a single line of code or conducting their first experiment — a principle covered extensively when building a defensible IP strategy as a scholar.

The situation becomes more complicated with collaborative projects spanning multiple institutions or involving private industry partners. In these cases, consortium agreements and sponsored research agreements typically pre-determine IP ownership and licensing rights. Failure to have these agreements in place before the research begins routinely leads to disputes that can freeze publication for years — a particularly damaging outcome in fast-moving fields like AI or genomics.

Attribution, Credit, and the Ethics-Law Gap

Authorship disputes represent a distinct but related problem. Copyright law protects expression, not ideas, and it recognizes joint authorship only when two or more people make independently copyrightable contributions to a single work. This creates a significant gap between legal authorship and academic credit norms. A senior professor who provides only high-level direction may receive first authorship under disciplinary convention while holding no legal copyright claim whatsoever. Understanding where ethical attribution ends and legal ownership begins is critical — and it directly connects to why plagiarism and copyright infringement are fundamentally different violations with different consequences and remedies.

Ghost authorship, gift authorship, and the omission of legitimate contributors are all recognized forms of research misconduct by bodies including the Committee on Publication Ethics (COPE) and the Office of Research Integrity (ORI). The ORI reported that authorship disputes featured in approximately 15% of its investigated misconduct cases between 2011 and 2021. Practically speaking, researchers should document contributions using frameworks like CRediT (Contributor Roles Taxonomy), which assigns 14 distinct contribution roles and is now required by over 50 major journals including all Elsevier and Springer Nature titles.

• Secure signed IP and authorship agreements before any collaborative work begins
• Maintain dated, witnessed lab notebooks — these serve as primary evidence in patent disputes
• Register copyrights for significant software outputs developed during research
• Verify whether grant terms (NIH, NSF, EU Horizon) impose open-access or march-in rights on resulting IP
• Negotiate publication rights explicitly in any industry-sponsored research contract

The academic IP environment rewards those who understand its mechanics before problems arise. Disputes over ownership and credit rarely resolve quickly or cheaply — the Stanford vs. Roche HIV test kit litigation ran for over a decade before reaching the Supreme Court in 2011. Prevention through clear agreements and meticulous documentation is not bureaucratic overhead; it is professional self-protection.

Pros and Cons of Copyright and Intellectual Property Management

Visual Content Protection: Copyright Enforcement Strategies for Digital Creators

Visual content is among the most frequently stolen intellectual property on the internet. A 2023 study by the Image Protection Alliance found that approximately 2.5 billion images are shared without proper attribution or licensing every single day. For photographers, illustrators, and graphic designers, this isn't an abstract legal issue — it directly translates to lost licensing revenue, diluted brand value, and unauthorized commercial exploitation of their creative work.

Proactive Protection Before Infringement Occurs

The most cost-effective enforcement strategy is prevention. Embedded metadata — specifically IPTC and XMP data fields — should contain your name, copyright notice, and licensing terms in every file before it leaves your possession. Many creators overlook this step, then find themselves unable to prove ownership when disputes arise. Tools like Adobe Bridge, ExifTool, or Lightroom's metadata templates can automate this process across entire photo libraries in minutes.

Visible watermarking remains controversial but effective. Semi-transparent watermarks positioned across the focal point of an image — not tucked in a corner where they're easily cropped — deter casual theft without making your portfolio unusable for previewing. For stock photographers on platforms like Shutterstock or Getty, the platform handles this automatically, but independent creators need to implement their own system. A common professional approach is to deliver watermark-free files only after payment confirmation via encrypted download links with expiration times of 24–48 hours.

Registering your work with the U.S. Copyright Office before publication gives you access to statutory damages of $750 to $150,000 per infringement, plus attorney's fees. Without registration, you're limited to actual damages — often nearly impossible to quantify for a single image. Bulk registration through the Electronic Copyright Office (eCO) system allows you to register up to 750 images as a single collection for a flat $65 fee, making it economically viable even for prolific creators.

Detection and Enforcement in Practice

Monitoring tools have matured significantly. Google Reverse Image Search, TinEye, and specialized services like Pixsy or Copytrack actively scan the web for unauthorized use of your images. Pixsy alone has processed over $20 million in settlements for photographers since its founding. These platforms operate on a contingency basis — typically taking 40–50% of any recovered damages — which means zero upfront cost for creators who lack legal budgets. When you understand how image theft actually happens across digital platforms, you can prioritize which detection tools to deploy based on where your work is most likely to appear without permission.

When infringement is confirmed, the standard escalation path moves through three stages: a DMCA takedown notice directly to the hosting platform (legally required to respond within 10–14 business days), a cease-and-desist letter with licensing demand, and finally civil litigation or small claims court for amounts under $10,000. Many creators successfully resolve cases at the first stage — major platforms like Instagram, Pinterest, and WordPress comply with DMCA requests reliably, and having documented proof of prior publication through timestamped posts or metadata history makes your case unambiguous.

• File copyright registration within 90 days of publication to maintain statutory damage eligibility
• Set up automated Google Alerts combining your name with terms like "photo credit" or "image courtesy"
• Keep raw files and original exports as proof of authorship — metadata alone can be stripped
• Document every infringement with full-page screenshots showing the infringing URL and date
• Use Copytrack's free infringement scanner monthly for any commercially valuable image catalog

Commercial licensing rates provide the benchmark for settlement negotiations. The Getty Images licensing calculator is widely accepted by courts as a reasonable market-rate reference — a single editorial image licensed for one year on a mid-traffic website typically commands $200–$500, giving you a concrete figure when drafting your initial demand letter without requiring an expensive appraisal.

Digital Plagiarism Detection Technologies: Tools, Algorithms, and Platform Comparisons

Modern plagiarism detection has evolved far beyond simple string-matching. Today's enterprise-grade tools use a combination of fingerprinting algorithms, semantic analysis, and machine learning models that can identify paraphrased content, translated text, and structurally rearranged passages with accuracy rates exceeding 95% in controlled benchmarks. Understanding what happens under the hood helps you choose the right tool for the right context — and avoid costly blind spots.

Core Detection Methodologies

The backbone of most detection platforms is shingling, a technique that breaks documents into overlapping token sequences (typically 5–10 words) and compares their hash values against a reference corpus. Turnitin's iThenticate, used widely in academic publishing, maintains a proprietary database of over 91 billion web pages plus 80 million academic papers — making it the industry benchmark for scholarly content. Copyscape, by contrast, focuses on web-published content and uses crawl-based indexing, which makes it effective for detecting duplicate marketing copy but weaker for unpublished manuscript theft.

Semantic similarity detection is where the field has advanced most dramatically since 2020. Tools like PlagScan and Unicheck now deploy transformer-based models (variants of BERT and its derivatives) that measure conceptual overlap rather than literal word matches. This matters enormously when dealing with the often-blurry boundary between inspiration and actual copying — a distinction that legal teams and content platforms must navigate daily. A passage rewritten at 70% word substitution may still trigger a semantic match score of 85% in these newer systems.

Visual and Multimedia Content Detection

Text-based tools are only part of the story. Perceptual hashing (pHash) and reverse image search APIs — such as those provided by Google Vision AI, TinEye, and Pixsy — analyze image content at the pixel and structural level, generating a unique fingerprint that remains stable even after resizing, cropping, or color adjustment. Pixsy alone has processed over 1.5 billion images and secured settlements in more than 50,000 infringement cases. If your work includes photographs, illustrations, or design assets, understanding how visual content gets copied and misappropriated across platforms is essential before selecting a monitoring solution.

For video and audio, Content ID on YouTube uses audio fingerprinting and frame-level analysis to match uploads against a rights-holder database containing over 800 million reference files. The system processes approximately 400 hours of video uploaded every minute — a scale no manual review process could approach. Similar ACR (Automatic Content Recognition) technology underpins platforms like Audible Magic and BMAT for broadcast monitoring.

Practical Platform Selection Criteria

When evaluating tools for your organization, prioritize these factors:

• Database scope: Does the platform index your specific content ecosystem — academic journals, web content, code repositories, or multimedia?
• API access: Enterprise workflows require programmatic integration; tools without robust APIs create manual bottlenecks at scale.
• False positive rate: Turnitin reports a false positive rate below 1% for academic text, but that figure climbs significantly for highly technical or formulaic content.
• Jurisdictional evidence packaging: Platforms like Copytrack and Pixsy auto-generate DMCA-compliant documentation, which directly reduces legal overhead.
• Multilingual support: Cross-language plagiarism detection remains technically immature; only a handful of tools handle translation-based copying with acceptable precision.

No single platform covers every content type or use case. The practical recommendation for most rights-holding organizations is a layered stack: one text-oriented tool with a deep academic or web corpus, one visual monitoring service with automated takedown workflows, and periodic manual audits for high-value assets that automated systems may miss.

Fair Use Doctrine and Creative Commons: Navigating Legal Reuse Frameworks

Two parallel systems govern legitimate content reuse in practice: the statutory fair use doctrine under U.S. copyright law (17 U.S.C. § 107) and the contractual Creative Commons licensing framework. Understanding how these systems interact — and where they diverge — is critical for anyone producing derivative works, educational materials, or commercial content. Neither framework is a blanket permission slip, and misreading either one carries real legal exposure.

The Four-Factor Fair Use Analysis

Courts evaluate fair use claims through four factors, none of which is individually decisive. The purpose and character of the use — particularly whether it is transformative — carries the most weight in modern jurisprudence following Campbell v. Acuff-Rose Music (1994). The nature of the copyrighted work matters less in most disputes, but using unpublished or highly creative works cuts against fair use. The amount and substantiality of the portion used is not purely quantitative: copying 300 words from a 100,000-word novel is factually minor, but copying the climactic plot twist is substantively significant. The market effect factor — whether the use harms the potential market for the original — is frequently the decisive issue in commercial contexts.

Practical thresholds taught in media law courses (e.g., "10% of a work is safe") have no statutory basis whatsoever. The Copyright Office explicitly warns against relying on such rules. Academic publishers like the Copyright Clearance Center process over 10 million licensing transactions annually precisely because institutions understand that educational purpose alone does not guarantee fair use protection. When working in academic contexts, understanding how scholars actively manage their own IP exposure puts fair use into sharper institutional relief.

Creative Commons: License Tiers and Common Misconceptions

Creative Commons licenses operate on a spectrum of six core combinations built from four elements: Attribution (BY), ShareAlike (SA), NonCommercial (NC), and NoDerivatives (ND). The most permissive, CC BY, requires only attribution and allows commercial use and adaptations. CC BY-NC-ND is the most restrictive and prohibits commercial use and any derivative works. A critical operational error is assuming "free to use" means "no conditions apply" — every CC license except CC0 (public domain dedication) requires attribution, and violating license conditions converts use into infringement.

Version matters more than most practitioners realize. CC 4.0 licenses (released 2013) are internationally harmonized and contain automatic license reinstatement provisions after inadvertent violations are cured. Earlier versions (3.0, 2.5) carry jurisdiction-specific language and no reinstatement clause. When licensing your own content or sourcing CC-licensed materials, always verify the version number in the license badge metadata.

• Check the license URL, not just the badge icon — badges are frequently misapplied or outdated
• NC licenses prohibit use in any revenue-generating context, including ad-supported websites
• SA licenses create a "viral" effect: derivative works must carry the same license, regardless of your original content's licensing
• CC licenses are irrevocable once granted, but apply only to specific versions of a work

Where fair use and Creative Commons intersect is a legally contested area. Using a CC-licensed work under fair use is technically possible but rarely necessary — if the license already permits the use, invoking fair use is redundant. However, a CC-licensed work used in ways that exceed the license's scope (e.g., commercial use under a NC license) may still qualify as fair use depending on the four-factor analysis. This distinction matters particularly when the lines between attribution failures and actual legal violations need to be drawn precisely, since license non-compliance and copyright infringement are not always the same wrong.

IP Risk Management in Collaborative Academic and Professional Environments

Collaboration is where intellectual property disputes are born. When two researchers co-author a paper, when a corporate team develops a new process, or when a university partners with an industry sponsor on applied research, the question of who owns what rarely gets answered before the work begins — and that silence becomes expensive. A 2022 survey by the Association of University Technology Managers found that IP disputes were the single most common reason research collaborations between universities and industry partners broke down before commercialization.

Establishing Ownership Before Work Begins

The most effective IP risk management happens before a single line of code is written or a single experiment is run. Joint ownership agreements, collaboration agreements, and contribution matrices are not bureaucratic formalities — they are the structural foundation that determines who can license, sell, or enforce rights later. In academic settings specifically, scholars often underestimate how institutional IP policies automatically assign ownership of work created using university resources, even when that work happens outside of normal working hours. Anyone working within an academic institution should closely examine their employment contract and the institution's IP policy before beginning any externally funded project. Practical guidance on how scholars and students can proactively assert and document their creative contributions is often the difference between retaining rights and losing them by default.

For professional environments, particularly in tech and pharma, invention assignment clauses in employment contracts typically assign all work-related inventions to the employer — sometimes including inventions made on personal time if they relate to the company's field of business. California is notably one of the few U.S. states with statutory limits on such clauses under Labor Code § 2870, but most jurisdictions offer no such protection. Teams should document the genesis of ideas, maintain timestamped lab notebooks or version-controlled repositories, and hold regular IP audits as projects evolve.

Navigating Third-Party Materials in Collaborative Work

Collaborative projects routinely incorporate third-party content — datasets, code libraries, images, published research — and this is where license incompatibility creates hidden liability. Mixing a dataset governed by a Creative Commons Attribution-NonCommercial license with commercially funded research can invalidate the entire project's commercial potential. Open-source software components under GPL licenses can impose copyleft obligations that extend to the entire codebase, a situation that has cost companies like Oracle and VMware millions in litigation costs and reputational damage.

One area that trips up even experienced researchers is the distinction between ethical misconduct and legal infringement. Understanding that plagiarism and copyright infringement operate under fundamentally different legal and ethical frameworks is critical when building compliance protocols for collaborative teams. A researcher can plagiarize without infringing copyright (copying ideas without attribution) and can infringe copyright without plagiarizing (reproducing licensed text with full attribution but without permission).

• Always execute a written collaboration agreement before any joint work begins, specifying ownership splits, licensing rights, and publication approval processes
• Audit all third-party inputs for license compatibility at project intake, not at publication or launch
• Maintain contribution logs with timestamps to establish authorship chains in case of later disputes
• Clarify sponsor rights in funded research — many industry sponsors negotiate for exclusive licenses or right-of-first-refusal clauses that can restrict academic publication timelines

The reality is that IP risk in collaborative environments is cumulative and often invisible until a commercialization or publication event forces the issue. Building IP hygiene into the workflow from day one — rather than treating it as a legal afterthought — is the operational standard that separates high-functioning research teams from those that spend their final months in arbitration instead of publication.

Cross-Border Copyright Enforcement: International IP Law and Jurisdictional Challenges

Copyright enforcement across national borders remains one of the most operationally complex challenges rights holders face. While the Berne Convention — currently ratified by 181 countries — establishes a foundational principle of reciprocal protection, the practical reality of enforcing those rights when infringement occurs in a foreign jurisdiction is a different matter entirely. A German photographer whose work is scraped by a platform hosted in Vietnam, or a U.S. software company whose code appears on servers in Eastern Europe, faces a labyrinth of conflicting national laws, inconsistent enforcement mechanisms, and jurisdictional ambiguity that can render even valid legal claims practically unenforceable.

The core problem is that copyright is territorial by nature. Your U.S. copyright registration, while valuable domestically, does not create an automatic enforcement mechanism abroad. What it does provide is evidentiary weight under treaties like the Berne Convention and the TRIPS Agreement (administered by the WTO), which obligate member states to provide copyright protection equivalent to their domestic standards. In practice, enforcement depends heavily on the legal infrastructure, political will, and judicial efficiency of the country where the infringement occurs. Countries like China, Ukraine, and several Southeast Asian nations consistently appear on the USTR's Priority Watch List precisely because treaty obligations and domestic enforcement remain misaligned.

Jurisdiction, Choice of Law, and the Internet Problem

Digital distribution has shattered the already fragile concept of territorial jurisdiction. When a work is uploaded in Brazil, hosted on servers in the Netherlands, and downloaded by users in Canada, determining which country's copyright law applies — and which courts have jurisdiction — is genuinely contested legal territory. Courts in the EU have increasingly applied the country of origin principle alongside the lex loci protectionis doctrine, meaning the law of the country where protection is claimed. U.S. courts, by contrast, have sometimes asserted extraterritorial jurisdiction when infringement causes substantial domestic economic harm, as seen in cases involving American platforms and foreign defendants.

Understanding the conceptual distinctions that underpin these cases matters. When cross-border disputes arise, courts must often determine whether an act constitutes actionable infringement or merely an ethically problematic appropriation that falls outside copyright's legal scope — a distinction that varies significantly across jurisdictions. This is not merely academic: a use that constitutes clear copyright infringement under U.S. law may be permissible under a foreign jurisdiction's broader fair dealing provisions.

Practical Enforcement Strategies for Rights Holders

Given these structural constraints, rights holders need a layered enforcement strategy rather than reliance on a single legal mechanism. The most effective approaches combine:

• Proactive registration in key markets — particularly the U.S., EU, China, and any jurisdiction where commercial exploitation is planned
• DMCA takedowns and platform-level enforcement — major platforms like Google, Meta, and YouTube operate globally and respond to properly filed notices regardless of where the infringing content originates
• Customs recordation — both the U.S. CBP and EU customs authorities allow copyright owners to record works, enabling border seizure of infringing physical goods
• Bilateral legal counsel — retaining local IP attorneys in high-risk jurisdictions before enforcement becomes necessary, not after

Visual content creators face a particularly acute version of this challenge, since images are trivially easy to strip of metadata and redistribute across platforms. Protecting visual works from unauthorized cross-border distribution requires combining technical measures like steganographic watermarking with active monitoring through services such as Pixsy or ImageRights International, which operate across more than 100 countries and can pursue claims on a contingency basis. Waiting until infringement is discovered organically is simply not a viable strategy at scale.

AI-Generated Content and Emerging Copyright Conflicts: Ownership in the Algorithmic Age

The legal frameworks governing copyright were built around a simple premise: a human author creates an original work. Generative AI has shattered that premise entirely. When Midjourney renders a photorealistic image in seconds, or GPT-4 drafts a 3,000-word article, the question of who—or what—holds the copyright becomes genuinely unanswered territory. Courts, legislators, and IP attorneys worldwide are scrambling to catch up with technology that moves faster than doctrine.

The U.S. Copyright Office has issued the clearest signal so far: purely AI-generated content, without sufficient human creative input, is not eligible for copyright protection. In the Thaler v. Perlmutter case (2023), a federal court upheld the Office's refusal to register an AI-generated image, reaffirming that copyright requires human authorship. However, the nuance matters enormously—if a human provides detailed prompts, makes curatorial selections, or substantially modifies AI output, some degree of copyright protection may attach to those human-authored elements.

The Training Data Problem: Whose Content Fuels the Machine?

The litigation landscape around training data is arguably more consequential than the authorship debate. Getty Images sued Stability AI in both the U.S. and UK for scraping over 12 million licensed photographs without compensation. The New York Times filed suit against OpenAI in late 2023, claiming verbatim reproduction of copyrighted journalism in model outputs. These cases will define whether ingesting copyrighted works for training constitutes fair use or systematic infringement at scale. The outcome could force a fundamental restructuring of how AI models are built and licensed.

For content creators, the parallel risk runs in both directions. AI tools trained on visual content can reproduce stylistically similar work that doesn't technically copy but closely mimics protected expression—a gray zone that resonates with anyone who has dealt with unauthorized reproduction of original visual work. The distinction between inspiration, imitation, and infringement becomes increasingly blurred when an algorithm can analyze millions of images and synthesize their patterns on demand.

Practical Risk Management for Creators and Enterprises

Organizations deploying generative AI need structured policies, not ad hoc responses. The key areas requiring immediate attention include:

• Input controls: Never feed proprietary, confidential, or third-party copyrighted material into AI systems without verifying the platform's data-use terms
• Output documentation: Maintain records of prompts, human edits, and creative decisions to substantiate any future copyright claims over AI-assisted work
• Indemnification clauses: Evaluate AI vendors' contractual protections—Microsoft, Adobe, and Google now offer limited IP indemnification for enterprise users of their AI tools
• Attribution hygiene: Clearly label AI-generated content, particularly in academic and journalistic contexts where safeguarding original scholarly contributions is paramount

One persistent misconception worth addressing directly: AI-generated content being ineligible for copyright does not mean it's free to use without consequence. It enters the public domain by default, meaning competitors can copy it freely—a significant commercial liability that many organizations haven't fully internalized. There is also a meaningful distinction between copyright infringement and plagiarism that affects how these disputes unfold; understanding where copyright law ends and ethical attribution begins is essential for anyone publishing AI-assisted work.

The consensus among IP specialists is that legislation will eventually impose disclosure requirements, licensing obligations for training data, and possibly neighboring rights frameworks similar to those protecting press publishers in the EU. Until that clarity arrives, the most defensible position is treating AI as a tool that augments human creativity rather than replaces it—and documenting every step of that creative process accordingly.