Provenance: Proving That Your Code Is Really Yours (medium.com)

A recent project delved into the complexities of establishing code ownership and copyright in the context of using large language models (LLMs). The experimental research highlighted the need for robust evidentiary frameworks like cryptographic provenance and watermarking to verify code originality. A key finding revealed that no single strategy is sufficient; instead, a combination of approaches is necessary to withstand adversarial scrutiny. Using various LLMs, the author tested their responses to proprietary code requests and found significant variability in how the models handled licensing information. Only one model flagged a proprietary license, raising concerns about the potential risk of unauthorized code adaptation. To address these issues, the author created an open-source command-line tool called Provenance, designed to generate cryptographic, timestamped records of code files. This tool provides a way to establish verifiable ownership without attempting to alter LLM behavior directly. Provenance includes functionalities to insert license headers, check for compliance, and create a cryptographic manifest of the codebase, all anchored to trustworthy time-stamping methods. This development signifies a crucial step toward safeguarding intellectual property in increasingly automated coding environments, highlighting both the ethical considerations and technical challenges faced by developers relying on LLMs.