Development dilemma: Cheap replication of AI models is raising questions

OpenAI, Anthropic PBC, and Alphabet Inc’s Google are reported to be working together to tackle the problem of unauthorised replication of their artificial-intelligence (AI) systems. They are coordinating through the Frontier Model Forum, a non-profit industrial body originally set up with Microsoft to promote the safe and responsible development of advanced AI models. OpenAI and Google Gemini are banning accounts that violate terms of service and proactively removing users who appear to be attempting to distil models through obfuscated third-party routers. However, the issue at hand is not just technological misuse but a deeper shift in the economics and geopolitics of AI, one that has direct implications for countries like India. At the centre of the debate is distillation, a machine-learning compression technique that allows a compact “student” AI model to be trained to replicate the behaviour and performance of a larger, complex “teacher” model. It retains high accuracy while reducing model size, speeding up inference, and lowering computational costs. But when used without authorisation to replicate proprietary systems, it becomes what the big tech firms call adversarial distillation. For the top generative AI companies based in the United States (US), this is a form of free-riding, allowing competitors to build similar products without bearing the massive costs of innovation.

 

The concern has intensified with the rise of DeepSeek and other high-performance Chinese models at significantly lower prices. Much of this advantage comes from an open-weight approach, where models are more accessible and easier to deploy. In contrast, US firms rely primarily on closed, proprietary systems that monetise access. The result is a growing divide in the global AI ecosystem. For India, this divide presents both an opportunity and a dilemma. On the one hand, cheaper and more accessible AI models can lower entry barriers for Indian startups, researchers, and public institutions. Historically, technology diffusion has driven broader economic gains, including fostering competition and accelerating innovation. In a country where digital public infrastructure has already transformed sectors from payments to governance, affordable AI could drive innovation in areas such as agriculture, education, and health care.

 

On the other hand, the concerns raised by US firms cannot be dismissed. If widespread replication undermines incentives to invest in frontier AI, it could slow the pace of cutting-edge innovation globally. India, which is still building its capabilities in advanced AI, depends in part on the continued progress of global frontier firms. Moreover, the security risks associated with replicated models, particularly those lacking safety guardrails, are equally relevant, where the misuse of AI could amplify challenges such as misinformation or cyber threats. This has allowed firms to frame adversarial distillation not just as a commercial risk but as a national-security concern. The tradeoffs must be considered. Protecting intellectual property is essential to sustain investment. But overprotecting it risks entrenching dominance and slowing the spread of innovation. Hence, a clearer global framework is needed to distinguish fair use from predatory extraction. In this context, India needs to form its own position, one that balances openness with safeguards, and access with accountability.