The Tech Docket Daily signal on tech & AI — what the world is searching, and why
Policy & Society

AI-Generated Code: Why Critics Say It Must Be Open Source

Exploring the debate on why AI-generated code should default to open source, legal precedents in the US and India, and local developer costs.

The debate over whether AI-generated code must default to open-source licensing has intensified following developer Jakub Jankiewicz’s provocative thesis. As courts globally rule that purely machine-authored outputs lack copyright protection, software teams face a choice: voluntarily dedicate AI-assisted code to the public domain or risk legal uncertainty over proprietary codebases built on scraped training data.

The Core Provocation: Why AI Code Demands Open Licensing

On November 15, 2026, developer Jakub T. Jankiewicz published an essay arguing that all code generated by artificial intelligence must default to open-source licenses. This demand aims to rectify the training debt incurred when frontier models scraped copyleft repositories without proper attribution.

Jankiewicz’s proposal struck a chord in the developer community, quickly rising to the front page of Hacker News on November 15, 2026, where it amassed more than 1,110 points and sparked over 590 comments explainx.ai. The core of the argument rests on “training debt” or “license laundering.” Frontier AI models were trained by scraping billions of lines of open-source code (including copyleft repositories and platforms like Stack Overflow) without preserving the required attribution or licensing terms jakub.jankiewicz.org. Critics argue that turning this collective, attribution-locked knowledge into proprietary software is an ethical violation that can only be resolved by releasing the resulting code under open-source terms or Creative Commons Zero (CC0) public domain dedications.

The concept of copyleft—a licensing practice that offers the right to freely distribute and modify software on the condition that the same rights are preserved in derivative works—is central to this debate. When an AI model ingests a repository licensed under the GNU General Public License (GPL) and subsequently outputs functionally identical or heavily derived code, it bypasses the license’s copyleft requirements. Jankiewicz argues that because these models could not exist without the open-source ecosystem, the software they generate should naturally belong to that same ecosystem.

United States courts have consistently ruled that purely AI-generated outputs are ineligible for copyright protection due to a lack of human authorship. This legal baseline forces developers using autonomous agents to either prove substantial human modification or accept that their code sits in the public domain.

On March 18, 2025, the U.S. Court of Appeals for the D.C. Circuit affirmed the U.S. Copyright Office’s denial of Stephen Thaler’s registration for an AI-generated artwork law.justia.com. Subsequently, on March 2, 2026, the Supreme Court of the United States denied certiorari, cementing human authorship as a bedrock requirement bakerdonelson.com. Meanwhile, the Ninth Circuit Court of Appeals addressed licensing liabilities on September 16, 2026, by affirming the dismissal of claims under DMCA Section 1202(b) in the Doe v. GitHub class action authorsalliance.org.

The Ninth Circuit panel held that generative AI tools do not “remove” or “alter” Copyright Management Information (CMI) when generating code because they probabilistically create new works rather than copying and stripping notices from stored files gibsondunn.com. However, this ruling did not resolve ordinary copyright infringement claims or pending breach-of-contract claims regarding the use of training data fenwick.com. Consequently, while AI providers are largely shielded from statutory DMCA penalties, the underlying code generated by their systems remains in a legal gray zone where its proprietary status is highly vulnerable to challenges.

The India angle

India’s legal framework offers a unique statutory path by attributing authorship of computer-generated works to the person who causes them to be created. However, recent rulings from the Indian Copyright Office and the Delhi High Court constrain this protection, forcing developers to navigate complex compliance boundaries.

Under Section 2(d)(vi) of India’s Copyright Act, 1957, the “author” of a computer-generated work is defined as “the person who causes the work to be created” ksandk.com. This provision, introduced via a 1994 amendment, originally contemplated computer-assisted creation rather than autonomous generative AI. In practice, this means that while Indian law does not outright exclude AI-assisted works from protection, the copyright itself must still vest in a human operator rather than the machine.

However, the originality standard remains a high hurdle. Under the Indian Supreme Court’s landmark ruling in Eastern Book Company v. D.B. Modak (2008), copyright only subsists if a work reflects a “modicum of creativity” born of human skill and judgment vintagelegalvl.com. Code generated by entering a simple prompt into an AI model with minimal human modification is unlikely to meet this threshold, rendering it public domain.

In September 2026, the Indian Copyright Office (ICO) officially rejected Stephen Thaler’s application to register the AI-generated artwork A Recent Entrance to Paradise with the AI system “DABUS” named as the author intellectual-property-helpdesk.ec.europa.eu. The ICO held that while computer-generated works can be original, authorship must vest in a human who caused the work to be created, rather than the machine itself surrey.ac.uk. This ruling aligns India’s administrative practice with the U.S. human-authorship requirement, clarifying that machines cannot hold intellectual property rights.

In the arena of landmark litigation, the Delhi High Court delivered India’s first substantive ruling on AI training and copyright in ANI Media Pvt. Ltd. v. OpenAI OpCo LLC. As explored in our coverage of the Delhi High Court’s fair dealing decision, Indian jurisprudence has begun actively distinguishing intermediate training from infringing public reproduction. The Court held, prima facie, that storing copyrighted works to train large language models qualifies as “fair dealing” under Section 52(1)(a)(i) of the Indian Copyright Act, 1957, as long as the use is limited to training and does not result in verbatim public reproduction spicyip.com.

Compliance with data privacy laws adds another layer of complexity for AI developers in the region. The Digital Personal Data Protection (DPDP) Act, 2023, along with its rules notified on November 13, 2025, requires all organizations in India to achieve full compliance by May 13, 2027 barandbench.com. Under this framework, using scraped datasets containing personally identifiable information (PII) for AI training without explicit consent can trigger statutory penalties scaling up to ₹250 crore per breach category khuranaandkhurana.com.

For individual developers and startups in India, utilizing premium AI coding tools also introduces a significant financial premium due to foreign exchange markups and local taxes. While a subscription to standard developer tools often costs a baseline rate of $10.00 per month, an Indian developer paying with a domestic credit card faces a much higher true landed cost. Assuming a baseline exchange rate of ₹95.00 per USD, the raw cost is ₹950.00, but adding a typical 3.5% bank forex markup (₹33.25) and India’s mandatory 18% Integrated GST on imported digital services (₹171.00) pushes the final monthly invoice to ₹1,154.25. This represents a true brand markup of 21.5% over the nominal exchange rate, making open-source, local-first tools like BharatCode highly attractive github.com.

This local-first approach has driven the rise of sovereign Indian open-source coding assistants. BharatCode is an open-source, local-first, Rust-powered terminal AI coding agent built specifically for Indian developers and enterprises github.com. This mirrors a broader global movement where developers are swapping terminal agents to gain granular control over codebases and API spend, avoiding foreign cloud egress.

Global jurisdictions have taken sharply divergent paths regarding the copyrightability of computer-generated code. The table below outlines how key international statutory frameworks treat purely AI-generated software:

Jurisdiction Primary Statute Authorship Rule Pure AI-Generated Code Status
United States 17 U.S.C. § 102 Strict human authorship required law.justia.com Public Domain (Uncopyrightable)
United Kingdom CDPA 1988 Sec. 9(3) Person making necessary arrangements halvertonandco.com Protected (50-year term)
India Copyright Act Sec. 2(d)(vi) Person who causes creation ksandk.com Protected if human causes and meets originality standard
European Union EU AI Act / InfoSoc Directive Intellectual creation of a human author europa.eu Generally Public Domain without human selection

This fragmentation means a software product developed globally could see its core code protected in London and New Delhi, but entirely uncopyrightable in New York or San Francisco. For multinational enterprises, this legal asymmetry creates an operational bottleneck, forcing legal teams to audit the precise human-to-machine ratio of their engineering pipelines.

Our take

Enforcing a mandatory open-source license on all AI-generated code is practically impossible due to the hybrid nature of modern software development. While the “license laundering” critique is ethically sound, the lack of a reliable line-by-line authorship metadata layer makes unilateral copyleft enforcement unworkable.

Modern software development is rarely a binary choice between pure human writing and pure autonomous generation. Instead, it is a fluid spectrum involving IDE autocompletion, agentic refactoring, and multi-file scaffolding. There is currently no standardized technical metadata layer (such as tagging at the Abstract Syntax Tree, or AST, level) that can reliably separate human keystrokes from machine completions. Attempting to enforce a rule where “any code touched by AI must be open source” would effectively force almost all modern commercial software to open-source its proprietary IP, given the near-ubiquitous adoption of coding assistants.

Furthermore, the ethical argument regarding training debt overlooks the fundamental nature of human learning. Human developers also learn to program by reading open-source, GPL-licensed repositories. When a human programmer writes proprietary code after studying copyleft examples, they do not violate the GPL unless they copy-paste code verbatim. This training debt is a direct consequence of the massive data ingestion pipelines that power frontier models, a process that has become so resource-intensive that even industry giants are adjusting their timelines, as detailed in our analysis of how OpenAI Slows Down AI Training.

The economic pressure to protect proprietary intellectual property is immense, given the massive capital flowing into AI hardware and infrastructure, as analyzed in our report on Nvidia’s circular demand engine. Rather than relying on unenforceable licensing mandates, the industry is more likely to settle on a compromise: robust indemnification clauses from AI vendors, coupled with strict enterprise code-filtering tools that prevent the verbatim generation of copyrighted training data. For deeper discussions on the intersection of artificial intelligence and public policy, visit our dedicated Policy & Society hub.

Frequently asked questions

Can I copyright code generated by ChatGPT?

Under current U.S. law, purely AI-generated code cannot be copyrighted because it lacks human authorship. However, if you write code in collaboration with an AI and contribute significant human intellectual labor, the human-authored portions may be eligible for protection.

How does the Indian Copyright Act handle AI-generated code?

Section 2(d)(vi) of India's Copyright Act, 1957, attributes authorship of computer-generated works to the person who causes the work to be created. However, the work must still demonstrate a 'modicum of creativity' born of human skill and judgment to be copyrightable.

Are there open-source AI coding tools available in India?

Yes, Indian developers have created prominent open-source alternatives like BharatCode, a Rust-powered terminal AI agent that runs locally, and Devika, an agentic software engineer. These tools allow developers to maintain data residency and manage API costs in Indian Rupees.

Sources & further reading

  1. All AI-code should be Open Source — Jankiewicz (primary source)
  2. Thaler v. Perlmutter — Justia (primary source)
  3. BharatCode CLI — GitHub
How this article was made: topic selected from same-day search-trend and community-momentum data across India and the US; researched, drafted and fact-checked with AI assistance under the site's automated quality gates (source citations, originality, no-clickbait and accuracy checks), on the editorial standards set by Saurab Jain. Details in our editorial policy. Spotted an error? Email a correction.

More of today, in 60 seconds: Today's Docket →