OPINION
This week, I came across research from an AI developer warning about capabilities that could potentially be used in dangerous ways.
And I found myself wondering: Who bears the responsibility when the people warning us about the risks are also the people developing the technology?
Imagine I invite guests to dinner and prepare food loaded with extremely hot chilli peppers. Then, as everyone sits down to eat, I announce:
“Be careful. This could burn your mouth.”
My guests might reasonably ask: If you knew that, why did you put so much chilli pepper in the food in the first place?
Perhaps that oversimplifies AI development, but I believe the underlying question deserves attention.
We have regulatory systems for potentially dangerous technologies. Aircraft, medicines and other complex products can undergo lengthy development, testing and regulatory review.
AI presents a different challenge.
Its capabilities are developing extraordinarily quickly. By the time governments, regulators and the public understand one generation of technology, developers may already be working on the next.
That makes it difficult for society to rely solely on regulation to keep pace.
In my view, this places an unusually large responsibility on the organizations developing advanced AI systems. Researching potential dangers and warning the public about them is important. But warnings should not become a substitute for asking whether certain capabilities should be developed, constrained or released in the first place.
This does not mean stopping AI development. AI is already demonstrating tremendous potential to help people learn, work, create and solve problems.
Nor should legitimate safety research be turned into frightening predictions about machines taking over the world.
There is an important difference between what researchers have demonstrated, what they believe could happen, and what others speculate might happen.
The public deserves to know that difference.
As AI becomes more capable, perhaps the question shouldn’t simply be:
“What can we build?”
We should also be asking:
“If we build it, what responsibility do we have for what happens next?”
Source context: Anthropic research on AI models and autonomous self-replication, October 12, 2026.