arXiv: Multi-Channel Spread-Spectrum Code Watermarking

This publication introduces a novel technical method for embedding invisible, tamper-resistant watermarks into AI-generated code using multi-channel spread-spectrum techniques. While not a regulatory change itself, it represents a significant advancement in AI safety and provenance tracking, directly relevant to the EU AI Act’s requirements for transparency and content authentication. The paper demonstrates how developers can mark code outputs from large language models to verify origin and detect unauthorized modification, addressing key compliance challenges around model accountability and misuse prevention.

Organizations developing or deploying generative AI systems for software development are most affected, particularly those in regulated sectors such as finance, healthcare, and critical infrastructure. Compliance teams in AI providers, cloud platforms, and software vendors should monitor this technique as a potential best practice for meeting Article 50 obligations on transparency and Article 55 on systemic risk mitigation. The method could also support audit trails required under the AI Act’s high-risk classification framework.

Compliance teams should immediately assess whether their current code generation pipelines include any form of watermarking or provenance tracking. If not, they should initiate a technical review of this and similar methods, engage with engineering teams to evaluate integration feasibility, and begin documenting how such techniques could satisfy upcoming conformity assessment requirements. Proactive adoption of watermarking may reduce future regulatory friction and demonstrate good-faith compliance efforts.