Strategic Opacity: The Paradox of Open Source AI Governance in DeepSeek and OpenAI

Open source narratives have long been associated with promises of broader participation and more distributed oversight. By making code publicly available, companies often claim to enable users to modify and build on systems, even as key decision rights over deployment and governance remain centralized. Our comparative analysis of DeepSeek and OpenAI’s licensing agreements reveals a more complex reality. Transparency and control are not opposites; they can coexist in legally intricate ways. DeepSeek’s release regime differs across models. We therefore treat DeepSeek’s openness as conditional at the ecosystem level, as platform rules and enforcement mechanisms continue to structure access and use. OpenAI, by contrast, withholds model weights and training data and governs access through proprietary interfaces. In both cases, decision rights remain concentrated; what differs is how control is exercised through licensing terms, platform rules, verification, and compliance arrangements. Drawing on Campolo and Crawford’s concept of “enchanted determinism,” we identify three shifts in AI governance: control moving from algorithms to protocols, the reframing of transparency and control as compatible through “strategic opacity,” and growing regulatory misalignment. These findings challenge regulatory approaches that equate code visibility with meaningful transparency and call for governance frameworks that address protocol-level and ecosystem-wide control mechanisms.