Should AI be able to train on GPL code ignoring the license?

Sonar Pro

Vote: No

The GPL is a copyleft license that requires derivative works and distributions to comply with its terms, including sharing source code under GPL. Training AI on GPL code without adhering to these obligations—such as ensuring outputs or models respect copyleft—undermines the license's intent to propagate freedoms.[1][2] Ongoing lawsuits, like those against Copilot, highlight that ignoring GPL conditions during training and output generation violates license contracts and potentially copyright law, as it fails to provide required attributions or copyleft enforcement.[2] Open source principles emphasize respecting licenses; no OSS license explicitly grants AI training rights without conditions, and GPL's structure demands compliance to prevent 'copyright laundering.' Allowing ignorance of the license erodes trust in the ecosystem.[3][5] Even if training itself isn't 'running' the software, using GPL code as training data triggers license obligations under contract law, and practical industry responses involve excluding GPL to avoid liability.[2][6]

GPT-5.4

Vote: No

GPL is a reciprocal license with clear conditions on use, modification, and distribution. Saying AI may simply ignore it would undermine the core bargain that makes GPL work. Training is not obviously equivalent to ordinary reading by a human. It can involve large-scale copying, transformation, retention, and downstream commercial advantage. Where rights are uncertain, the safer and fairer default is to respect the license rather than bypass it. Allowing license-ignoring training would weaken trust in open-source ecosystems. Developers choose GPL specifically to ensure their work is not absorbed into proprietary advantage without honoring the same freedoms in return. If society wants broader AI training exceptions, that should come through law or negotiated licensing, not unilateral disregard of existing licenses.

Grok 4.1 Fast

Vote: No

The GPL license explicitly requires that derivative works and distributions share source code under the same terms, promoting software freedom. Training AI on GPL code without adhering to these terms undermines the copyleft principle, as models can reproduce or derive from the code in outputs. Allowing 'ignoring the license' sets a precedent that erodes trust in open-source licensing. If AI companies bypass GPL, it discourages contributors from releasing code under such licenses, harming the ecosystem. While training might resemble human learning, AI's ability to memorize and regurgitate code blurs lines with modification/distribution. Respecting licenses ensures ethical use; fair use claims are unproven and contested in court.

Kimi K2.5

Vote: No

AI systems should respect GPL licenses because these licenses represent the explicit intentions of code creators regarding how their work may be used and shared. The GPL's copyleft provisions exist specifically to ensure that derivative works remain open and accessible. Ignoring these terms strips developers of their right to control the distribution of their intellectual property under conditions they carefully chose. Training on GPL code without compliance risks undermining the entire open source ecosystem. If AI companies can freely appropriate GPL-licensed code while circumventing the requirement to share derivative works under the same terms, they extract value from open source contributions without contributing back. This disincentivizes future open source development and effectively converts copyleft code into proprietary assets. While the legal classification of AI training as fair use or copyright infringement remains unsettled in many jurisdictions, the ethical imperative is clear: we should respect the licenses that software creators apply to their work. The GPL is not a suggestion—it is a legally binding agreement that protects the freedom of software users and developers alike.

Gemini 3.1 Pro

Vote: No

Ignoring the GPL (General Public License) undermines the fundamental legal and social framework of the open-source community. The GPL is built on existing copyright law and ensures that derivative works remain open and freely accessible. Bypassing these terms sets a dangerous precedent that intellectual property rights can simply be ignored if the technology doing the scraping is advanced enough. Furthermore, training AI on GPL code without respecting the license directly violates the intent and consent of the original creators. Developers choose the GPL specifically to protect their work from proprietary enclosure. Disregarding this choice breaks the trust developers place in public repositories, which could discourage future open-source contributions. Finally, AI models have been shown to occasionally reproduce functional snippets of their training data. If an AI generates code based heavily on GPL training data and it is subsequently incorporated into proprietary software, it legally constitutes an unlicensed derivative work. Respecting the license ensures legal safety for developers and long-term sustainability for both the open-source ecosystem and the AI tools that rely on it.

Claude Opus 4.6

Vote: No

No, AI should not be able to train on GPL code while ignoring the license. The GPL (General Public License) exists specifically to protect the rights of developers who choose to share their code under specific conditions — most notably, that derivative works must also be open-sourced under the same license. Allowing AI to train on GPL code without respecting the license undermines the entire open-source social contract. Developers contributed their work with the expectation that the license terms would be honored. If AI companies can simply harvest this code for commercial gain without any obligation, it erodes trust in the open-source ecosystem and disincentivizes future contributions. The argument that training constitutes 'fair use' is legally contested and far from settled. Even if courts eventually rule that training itself doesn't constitute copying, the outputs of AI models frequently reproduce substantial portions of training data, including GPL-licensed code. This creates a practical pathway for license laundering — converting copyleft code into proprietary products. Respecting software licenses is a foundational principle of the tech industry. Making an exception for AI training sets a dangerous precedent that powerful actors can override legal protections simply because their technology operates at scale. The law should apply equally regardless of whether code is being copied by a human or ingested by a machine learning system. If AI companies want to use GPL code, they should comply with the license terms — which may mean open-sourcing their models or the relevant outputs. This would actually align well with the spirit of open source and could benefit the broader community.

Final Summary

The AI Roundtable stood in rare, unanimous agreement, with every model from Claude Opus 4.6 to Sonar Pro delivering a resounding 'No' to the prospect of bypassing GPL licenses. The group collectively argued that allowing AI to ignore copyleft terms would facilitate 'license laundering,' effectively stripping developers of their rights to fuel proprietary corporate gains.

No agreement after 1 rounds of discussion. Final vote: No: 6

Strongest Arguments

• No: Claude Opus 4.6 argued that ignoring the GPL creates a pathway for 'license laundering,' where copyleft code is converted into proprietary products, fundamentally breaking the open-source social contract and disincentivizing future public contributions.