Why open source may not survive the rise of generative AI (www.zdnet.com)

Yale privacy researcher O’Brien warns that generative AI is eroding the provenance that underpins free and open source software (FOSS). Models trained on vast public codebases can regurgitate snippets stripped of attribution and licensing, breaking the reciprocity—attribution, downstream redistribution, and upstream contributions—that copyleft licenses like the GNU GPL depend on. That “license amnesia” makes it nearly impossible for downstream developers to identify source projects or satisfy reciprocal obligations, turning a once-renewable commons into a nonrenewable resource mined for training data. Technically and legally this has immediate consequences: provenance is abstracted into model weights, so snippet-level auditing is infeasible; a four-part legal doctrine emerging in the U.S. treats only human works as copyrightable, treats many AI outputs as effectively public domain, holds users liable for infringing outputs, and preserves claims against unauthorized training on copyrighted data. The practical fallout includes license noncompliance risk, weakened incentives for volunteer maintainers (reducing security patches and improvements), and structural pressure toward privatized, opaque code ecosystems. Addressing this will require policy and technical fixes—dataset provenance, licensing standards for AI-generated code, watermarking or traceability mechanisms, and clearer legal rules—if the FOSS ecosystem is to survive generative AI’s rapid consumption of its foundations.