// wiki · analysis · September 2026

Vibe coding and lawyers

Draft article. The term “vibe coding” is colloquial; the legal questions below apply to any code created predominantly by generation.

The short answer

Vibe coding is creating programs in conversation with a model, where the person formulates tasks and accepts the result without writing code by hand. For the law these are two different questions. As a product: code without a provable human contribution is weakly protected, and its licensing cleanliness needs checking — the model was trained, among other things, on other people's open code. As an instrument: for a lawyer vibe coding is a fast way to assemble a prototype of a table, a bot or a digest, with the same verification rules as for texts.

What it is

The term settled in 2025 for a development style where the model generates code from a description and the “developer” runs the dialogue and tests the result. The product looks like ordinary software — from the users' and deals' point of view there is no difference; the difference appears at the question “whose is it and what is it based on”.

Rights to a vibe-coded product

• Authorship: the logic of case No. 02-4220/2025 carries over to code — a generation without a human's creative contribution is not protected; the contribution counts as architectural decisions, reworking, integration, not the mere fact of a request (details on generations).
• Service terms: rights to the model's output are set by the tool's user agreement — the first page a lawyer reads.
• Registration: a program with a sufficient formalised contribution is registered as software and enters the register — with evidentiary and tax value (the legal nature of AI products).
• Recording the contribution: keep the history of requests, edits, reviews — this is evidence for the future.

Licensing risks

The model could have reproduced fragments of open code from its training data — including copyleft. For a serious vibe-coded product the same instruments work as for any other: a component list (SBOM), scanners matching known open projects, the rule “GPL/AGPL — to review” (open source in the product). A separate risk is coincidence with others' patented solutions: generation does not exempt from a freedom-to-operate check.

Liability

For bugs in a vibe-coded product its owner answers, not the model: contract and consumer law give no discount for the way the code was written. A lawyer-creator should apply their own protocol: testing, verification of the data it produces, interface restrictions. A tool for internal tasks follows the same AI policy as texts (secrets and data).

How it helps lawyers

• Prototypes of instruments: condition tables, redline comparisons, digest bots — assembled in an evening without developers; on systematically building assistants — the academy workshop.
• Automating routine: mass extraction, drafts of letters, search over one's own archives in a closed contour.
• Limits: client data — only in the agreed contour; the result — as a draft verified by a human (the verification protocol).

Formalising rights to a vibe-coded product and building the check is the task of technology rights practice.