Unboxing Schrödinger’s Dataset: The real level of IP risk in AI generated code

AI is transforming software development, but for leaders in regulated industries, the legal risks related to AI-generated code, particularly intellectual property (IP) liability, pose significant challenges. The uncertainty about the training datasets of AI models complicates the integration of AI into development processes, with many CIOs expressing concerns about copyright infringement. A study from Carnegie Mellon University suggests that the actual risk of AI-generated code infringing on IP rights is considerably lower than feared, with occurrences of license-protected code generation being minimal. Tabnine offers a secure AI software development platform designed to address these concerns through inference-time and training-time protections, ensuring compliance without compromising performance or privacy. Their platform allows organizations to implement AI solutions while safeguarding against IP liabilities by using mechanisms like Provenance and Attribution, which check AI-generated code against publicly available code to ensure compliance with license standards. This approach enables enterprises to adopt AI with confidence, maintaining a balance between innovation and legal and security requirements.