To stop regulatory capture, we must move away from top-down mandates written by industry lobbyists. When standards require expensive, proprietary audits or massive compute-heavy testing, only a few trillion-dollar companies can comply. This creates a high barrier to entry that crushes startups. Instead, international bodies should favor open-source benchmarks and modular compliance frameworks. By using decentralized verification methods, smaller firms can prove their models are safe without needing a massive legal department or a mountain of cash.
We also need to bridge the gap for researchers in the Global South. Currently, most AI governance happens in Washington or Brussels. This ignores the unique socio-technical risks faced in developing nations. We should promote tiered regulation. Small-scale research projects or niche applications should not face the same heavy burdens as massive frontier models. If a startup in Nairobi develops a localized LLM for agriculture, they shouldn't need a global safety certification that costs millions. Instead, we should fund regional hubs that provide technical assistance and localized oversight. This keeps the ecosystem competitive and ensures safety rules actually reflect the realities of the people they are meant to protect.
