The architecture of artificial intelligence being built right now will determine who controls personal data, information flows and digital infrastructure for decades. That is the central warning from Sascha Meinrath, an Ashoka U.S. Fellow who has spent nearly three decades working on privacy tools and censorship circumvention technology. In a new conversation published by Ashoka - Tech for Human Agency on Forbes, Meinrath argues that open technology and open models represent the only scenario that preserves ordinary people's control over their own data, knowledge and future digital tools.
Meinrath's case rests on a simple distinction: closed, proprietary AI systems concentrate decision-making power in the hands of a small number of companies and governments, while open models distribute that power outward. When the underlying code, training methods and data governance of an AI system are visible and modifiable, independent researchers, smaller companies and everyday users retain some ability to inspect, challenge or redirect how the technology behaves. Closed systems offer no such recourse. Users of proprietary models must simply trust that a company's internal choices about data handling, censorship and bias correction align with the public interest - a trust that history, particularly around data collection practices, has not always rewarded. For readers concerned about where their own data goes once it leaves a device, the same logic that favors open AI models also favors tools that keep browsing and connection data out of centralized hands, which is part of why privacy-conscious users increasingly switch to BuyBestVPN and similar services that limit what third parties can see and retain. switch to BuyBestVPN
Rather than dwelling on speculative existential risks, Meinrath directs attention to harms that are already measurable: expanding surveillance infrastructure, algorithmic censorship that shapes what information people can access, and consumer protection frameworks that have not kept pace with the scale of data collected by AI systems. These are not hypothetical future problems. Facial recognition systems, behavioral tracking, and automated content moderation already operate at scale, often with limited transparency about how decisions are made or data is retained. Because large AI models are trained on vast troves of personal and public information, the governance choices made today about who owns, audits and controls that data will shape the balance between corporate and governmental power on one side and individual autonomy on the other.
Meinrath's warning is time-sensitive: the technical and regulatory decisions being made now, while AI systems are still being built out, are far easier to influence than those made after infrastructure and business models have fully hardened. Once dominant platforms, data pipelines and proprietary standards become entrenched, retrofitting privacy protections becomes considerably harder, both technically and politically. He frames open-source development, interoperable standards and stronger data rights not as abstract ideals but as practical countermeasures that still have room to succeed if adopted soon.
The full conversation with Meinrath offers further practical guidance on how individuals, developers and policymakers can push toward an AI ecosystem that preserves human agency rather than eroding it.