--- title: "Who Owns the Code in a Custom AI Development?" description: "Who owns the code in a custom AI development? Understand IP rights, licensing traps, and what to demand in your contract before signing anything." slug: "who-owns-the-code-in-custom-ai-development" url: "https://catalizadora.ai/blog/who-owns-the-code-in-custom-ai-development" cluster: "roi-ia-decision" author: "Pablo Estrada" published_at: "2026-06-20T10:35:25.519+00:00" updated_at: "2026-06-20T10:35:25.569831+00:00" read_minutes: "8" lang: "en" --- # Who Owns the Code in a Custom AI Development? > Who owns the code in a custom AI development? Understand IP rights, licensing traps, and what to demand in your contract before signing anything. # Who Owns the Code in a Custom AI Development? Sign the wrong AI development contract and you could end up paying recurring license fees—forever—for software your own team conceived, specified, and funded. The question of **who owns the code in a custom AI development** project is not a formality. It is a financial and strategic decision that affects your company's valuation, your ability to pivot, and your total cost of ownership over a multi-year horizon. This article gives you a precise, clause-level answer. --- ## Why Code Ownership in AI Projects Is More Complicated Than in Traditional Software A typical custom web application has one clear deliverable: source code. AI-native software adds layers: - **Trained model weights** — the numerical parameters learned during training, which can represent millions of dollars of compute and data. - **Fine-tuning datasets** — curated, labeled data your company provided or paid to create. - **Prompt libraries and chains** — the logic that orchestrates LLM calls, often the real competitive advantage. - **Inference infrastructure configuration** — Docker images, cloud architecture, CI/CD pipelines. - **Evaluation frameworks** — the test suites that measure model performance against your specific KPIs. Each of these components can be owned separately. A vendor can hand you the application code while retaining the model weights, or license you the prompt layer while keeping the training data pipeline proprietary. Without explicit contract language covering **all five layers**, you do not own the complete system. --- ## The Three Most Common Ownership Structures ### 1. Full IP Transfer (Work-for-Hire) Under a work-for-hire agreement, every artifact produced—code, weights, datasets, documentation—transfers to the client upon payment. This is the gold standard for companies that intend to: - Resell or white-label the software. - Raise venture capital (investors scrutinize IP ownership in due diligence). - Integrate the system into a future acquisition or merger. - Modify and extend the product without vendor approval. **What to look for in the contract:** Language stating that all deliverables are "works made for hire" under applicable copyright law, with an explicit IP assignment clause as a fallback in jurisdictions where work-for-hire doctrine is narrow. ### 2. Exclusive License The vendor retains copyright but grants the client an exclusive, perpetual, irrevocable license. The client can use, modify, and sublicense—but cannot claim ownership for M&A or IP monetization purposes. This structure is common with vendors who build on proprietary internal frameworks they do not want to transfer. **Risk:** "Exclusive" licenses have edge cases. What happens if the vendor is acquired? Does exclusivity survive the acquirer's terms? Always add a clause specifying that the license survives any change of control of the vendor. ### 3. SaaS / Subscription Model Disguised as Custom Development This is the structure that catches most buyers off guard. The vendor builds a system that appears custom but runs on their proprietary platform. You get a configured instance, not owned software. When you stop paying, access ends. Red flags: - Monthly or annual fees that persist after delivery. - No source code delivery milestone in the project timeline. - Deployment happens exclusively on the vendor's infrastructure. - Contract terms reference "platform access" rather than "software delivery." --- ## Who Owns the Code in a Custom AI Development: What the Law Says Copyright law in the United States and most LATAM jurisdictions defaults to the **creator** as the owner. If a contractor writes code for you without a written agreement, they own it—not you. The client's payment covers the service, not the copyright, unless the contract says otherwise. Key legal anchors: - **U.S. Copyright Act, 17 U.S.C. § 101**: Work-for-hire requires either an employment relationship or a written agreement for specific categories of commissioned works. Software is not automatically in those categories, making an explicit IP assignment clause essential. - **Mexico's Federal Copyright Law (Ley Federal del Derecho de Autor), Art. 84**: Commissioned works belong to the author unless a written contract specifies otherwise. - **Colombia's Law 23 of 1982, Art. 20**: Software created under an employment contract belongs to the employer, but independent contractors retain rights without a written assignment. **Bottom line:** In every jurisdiction, a written IP assignment clause is not optional—it is the only reliable protection. --- ## The Hidden Cost of Not Owning Your AI Code Consider a concrete scenario: a mid-sized logistics company in Mexico spends $180,000 building a route-optimization AI with a vendor who retains the model weights. Eighteen months later, the vendor raises prices by 40% or pivots their product. The client's options are: 1. Pay the new price. 2. Rebuild from scratch, losing 18 months of training data and iteration. 3. Negotiate a buyout of the IP—at the vendor's price. None of these are good outcomes. The $180,000 investment now has a recurring cost attached to it indefinitely, and the company cannot use the system as a proprietary asset on its balance sheet. Contrast this with full IP ownership: the same $180,000 creates a depreciable asset, a defensible competitive moat, and the freedom to hire any engineer to maintain or extend it. --- ## What to Demand in Your Custom AI Development Contract Before signing any custom AI engagement, validate these five points: **1. Complete IP assignment clause** All deliverables—source code, model weights, datasets, prompts, documentation—assign to the client upon full payment. No carve-outs for "pre-existing IP" embedded in the deliverables without explicit identification upfront. **2. Source code escrow or direct delivery** The source code must be delivered to the client, not just deployed. Deployment without code delivery means you are renting, not owning. **3. No recurring license fees post-delivery** If the system requires ongoing cloud infrastructure (AWS, GCP, Azure), those costs are yours to control—not a markup the vendor routes through their billing. There should be zero vendor license fees after the project closes. **4. Data ownership** Any training data you provided, labeled, or funded belongs to you. The contract should explicitly prohibit the vendor from using it to train other clients' systems. **5. Change-of-control survival clause** If the vendor is acquired, merged, or dissolved, your license or ownership rights survive unchanged. --- ## How Ownership Structures Affect ROI Calculations When a CFO models the ROI of a custom AI project, ownership structure materially changes the math: | Structure | Year 1 Cost | Year 3 Cost | Balance Sheet Impact | |---|---|---|---| | Full IP Transfer | $150K (fixed) | $0 vendor fees | Depreciable asset | | Exclusive License | $150K + $15K/yr | $180K total | No asset recognition | | SaaS Platform | $80K setup + $40K/yr | $200K total | Expense, not asset | The SaaS/platform model often looks cheapest at signing. Over 36 months, it frequently costs more—and leaves the company with nothing if they cancel. --- ## Questions to Ask Every AI Development Vendor These seven questions will expose ownership gaps in any proposal: 1. "Will we receive all source code, including model training scripts, at delivery?" 2. "Do you use any proprietary frameworks or platforms that are not included in the IP transfer?" 3. "Are there any recurring fees after project delivery?" 4. "Can we deploy this system on our own cloud account, independently of your infrastructure?" 5. "What happens to our IP rights if your company is acquired?" 6. "Do your contracts include an explicit IP assignment clause, or only a license?" 7. "Can we share your standard contract with our legal team before the discovery call?" A vendor that hesitates on any of these questions is telling you something important. --- ## What Full Ownership Looks Like in Practice At Catalizadora, every engagement—whether it's a 12-week **Core** build, a 15-day **Solo** sprint, or a scoped **Forge** project—ships with 100% IP and code ownership transferred to the client. That means: - Full source code delivery, including training pipelines and prompt architecture. - Zero recurring license fees. You pay for your own cloud infrastructure directly. - No proprietary platform lock-in. The system runs wherever you choose to host it. - A contract with an explicit IP assignment clause, reviewed by your legal team before work begins. This is not a differentiator we invented—it is what responsible custom development has always required. We just make it the default. --- ## Summary: The Code Ownership Checklist Before you sign any custom AI development contract, confirm: - [ ] Explicit IP assignment clause covering code, weights, data, and prompts. - [ ] Source code delivered directly to the client, not just deployed. - [ ] Zero vendor license fees after project delivery. - [ ] Training data ownership retained by the client. - [ ] Change-of-control survival language included. - [ ] Vendor answered all seven ownership questions without hesitation. --- ## Build Software You Actually Own The question of who owns the code in a custom AI development project has a clear answer: **you should**. Anything less is a liability dressed up as a service. If you want to understand how we structure ownership, timelines, and pricing before committing to a single line of code, read [our manifesto](/manifiesto)—it explains exactly how we think about building software that belongs to you. ## Preguntas frecuentes ### Who owns the code in a custom AI development project by default? Under U.S. and most LATAM copyright laws, the creator (the vendor or contractor) owns the code by default unless a written IP assignment or work-for-hire clause explicitly transfers ownership to the client. Never assume payment implies ownership. ### What is the difference between an IP assignment and a software license? An IP assignment transfers copyright ownership to you permanently. A license grants you permission to use the software under defined conditions—which can be revoked or expire. For full control, demand an IP assignment, not just a license. ### Do model weights count as code in an AI development contract? Model weights are a separate, highly valuable artifact from the source code. A contract must explicitly cover weights, training datasets, prompt libraries, and inference infrastructure—not just application source code—to provide complete protection. ### Can I use AI software built for me on my own cloud account? Only if the contract includes source code delivery and no platform lock-in clauses. Many vendors deploy on their own infrastructure, meaning you are renting access rather than owning the system. Always verify you can redeploy independently. ### What happens to my AI software if the development vendor is acquired? Without a change-of-control survival clause, your license or ownership rights could be renegotiated by the acquirer. Always include explicit language stating that your IP rights survive any change of control of the vendor. ### Are recurring fees after delivery a sign that I don't own the software? They can be. Recurring vendor fees after delivery often indicate a platform or SaaS model where you are renting access, not owning the software. Legitimate infrastructure costs (AWS, GCP, etc.) billed directly to you are normal; vendor license fees are a red flag. --- Source: https://catalizadora.ai/blog/who-owns-the-code-in-custom-ai-development Author: Pablo Estrada — AI Catalyst, LLC (catalizadora.ai)