LLMs struggle with code provenance and proprietary license headers

A developer's weekend project exploring LLM guardrails and code provenance revealed complexities in proving code ownership. The developer tested several frontier LLMs, including Gemini, OpenAI, Perplexity, and Claude Sonnet 5, to see how they handled proprietary license headers. While some models like Claude Sonnet 5 flagged the header and sought confirmation, others like Gemini converted the code without comment, raising questions about the effectiveness and consistency of current LLM safety measures in respecting intellectual property. AI

IMPACT Highlights potential inconsistencies in LLM adherence to intellectual property and licensing, impacting how developers can ethically use AI tools for code generation.

RANK_REASON Developer's personal project exploring LLM behavior and limitations.

Read on dev.to — LLM tag →

• arXiv
• Claude Sonnet 5
• Gemini
• OpenAI
• Perplexity
• VEKTOR Memory Pty Ltd

• product
• policy

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →