Noubar Afeyan, the billionaire co-founder and chairman of Moderna, wants AI agents regulated before they reach customers. His argument is that waiting until after something goes badly wrong is the wrong order of operations.
Afeyan made the case at the Bloomberg New Economy Forum in New Delhi on October 13, 2026. His chosen cautionary tale was thalidomide, a drug from the 1950s and 60s that caused birth defects because it was not adequately tested for safety.
Capability versus intentionality
The core of Afeyan’s pitch is a distinction most AI debates tend to blur. He separates large language models from AI agents and argues they deserve very different treatment.
Afeyan described LLMs as representing “capability.” A language model can write, summarize and reason, but it mostly waits for a person to tell it what to do.
AI agents, in his framing, represent “intentionality.” They take autonomous actions in the world, pursuing goals with less human supervision at each step.
That gap is where Afeyan thinks oversight should focus. His position targets the commercial deployment of agents, not the foundational models underneath them.
“I believe what we should do is to think carefully about regulating agents that are being commercialized.”
The qualifier “being commercialized” matters. Afeyan is not calling for a freeze on research or a clampdown on model development. He is urging industry leaders to build safety checks that happen before an agent goes on sale.
The upside and the hazards
Afeyan is not an AI skeptic. He acknowledged that agents hold real promise for speeding up drug discovery and pushing scientific research forward.
The same autonomy is what worries him. Afeyan flagged the risk that agents could generate harmful compounds, or produce unreliable results when they operate outside regulated environments.
A wider push on agent governance
Afeyan’s remarks land in the middle of a broader debate over how to govern agentic AI. That conversation has been gathering momentum across the industry throughout 2026.
Organizations including OpenAI, along with various international bodies, have launched their own initiatives on AI governance. What sets Afeyan’s intervention apart is its narrow target: the moment an agent becomes a commercial product.
That framing borrows directly from how medicine works. Drug developers are expected to show a product is safe before it is sold, rather than collecting complaints after the fact. Afeyan is essentially asking the AI sector to adopt the same instinct voluntarily, before regulators impose one.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.