Moderna Founder Urges Urgent Regulation of AI Agents Before the First Catastrophe

Noubar Afeyan, co-founder of Moderna and a biotech billionaire, has called for urgent regulation of AI agents, which tech companies are already selling to customers. His argument comes from his own industry's experience: if we wait for a catastrophe and write the rules afterward, we will repeat the history of the drug industry. Bloomberg reported this.

The piece is short. The available text does not say what exactly Afeyan proposes (which requirements, which regulator, what timeline). So below is only what is known for certain, plus an explanation of why the comparison to pharma is not accidental.

Who is speaking and why people will listen

Afeyan is a co-founder and chairman of Moderna, and the founder of the venture studio Flagship Pioneering. He knows what a product's path from lab to mass market looks like under the oversight of regulators such as the FDA. Some Flagship companies use AI in drug development. This is the voice of someone from a tightly regulated industry who also has a stake in the technology's growth.

What the history of pharma teaches

Pharmaceutical rules rarely arrived in advance. Usually a tragedy came first.

• 1937, US: the drug Elixir Sulfanilamide, whose solvent was toxic, killed more than a hundred people. A 1938 law followed, requiring manufacturers to prove safety before going to market.
• Late 1950s and early 1960s: thalidomide caused birth defects in thousands of children around the world. In the US, the response was the Kefauver-Harris Amendments of 1962, which added a requirement to prove a drug's efficacy.

In both cases, the oversight system was built in the wake of harm. According to Bloomberg's report, Afeyan proposes not waiting for the same scenario to play out with AI.

Why agents differ from chatbots

A chatbot produces text, and a human decides what to do with it. An agent acts on its own: it makes purchases, writes and runs code, sends emails, and gets access to corporate systems and accounts. Here a model error turns into an action that cannot always be undone. The more authority an agent gets, the higher the cost of a failure, and the harder it is to work out who is responsible: the model developer, the platform provider, or the company that deployed the product.

In most jurisdictions, there are still few requirements aimed specifically at agents. General frameworks for AI exist, but they were written for models and systems as a whole, not for programs that carry out chains of actions on their own.

Key facts

• Who: Noubar Afeyan, co-founder of Moderna, founder of Flagship Pioneering.
• What he demands: urgent regulation of AI agents sold by tech companies.
• Argument: waiting for a catastrophe will repeat pharma's mistakes.
• Source: Bloomberg, October 13, 2026.
• Not disclosed: specific measures, the regulator, and the timeline.

What companies already deploying agents should do

There is no need to wait for a law; pharma already offers practices that work.

First: limit permissions. An agent should get the minimum access it needs, and irreversible operations such as payments, data deletion, and sending external emails should require human confirmation.

Second: keep a log. Every agent action should be recorded so that, after a failure, the chain of decisions can be reconstructed.

Third: roll out in stages. By analogy with the phases of clinical trials, start with a narrow task and a limited group of users, and expand the agent's scope only after testing on real data.

Fourth: decide in advance who is liable for damage, and write it into contracts with vendors. If regulation arrives, companies with a ready audit trail and clear accountability will come out ahead.