i
News
News · 2026-09-28

Steven Pinker wants AI safety focused on real-world threats

@neuronium_ai @neuronium_ai

Steven Pinker is arguing that AI safety should focus less on visions of human extinction and more on threats that can be addressed through engineering and oversight. The Harvard psychologist says bioterrorism, cyberattacks and uncontrolled AI agents deserve serious attention—but apocalyptic language does little to prevent them. His intervention comes as he acknowledges that he underestimated how capable large language models would become and how recklessly companies would release products.

Cover: Steven Pinker wants AI safety focused on real-world threats

The risks Pinker takes seriously

Pinker rejects scenarios such as the “paperclip maximizer,” in which an AI sacrifices humanity to make paperclips. Such stories, he argues, confuse intelligence with a desire for dominance and assume a system would fixate maniacally on a single goal.

His alternative is not to dismiss AI risk. He points to bioterrorism, cyberattacks and AI agents operating without control as serious dangers. The response, in his view, should be practical safeguards built around:

Independent oversight
Accountability
Human control

Pinker has previously been optimistic about AI. He now says he misjudged how capable large language models would become, and did not expect companies to release their products so recklessly. There is an irony in the race, he notes: it is being led by people from the same community warning about these risks.

What safety rhetoric leaves out

In June, Alexander laid out a different position in a blog post. He puts the chance that AI will destroy humanity at 20 percent and calls for more safety research and international agreements to slow AI development.

I think the disagreement is less about whether AI poses risks than about which risks should set the agenda. Pinker’s case is strongest when it replaces an abstract extinction story with specific hazards and safeguards. But calling for oversight and accountability is not the same as showing how to make them work across companies building systems at speed.

The unresolved tension is practical: safety measures need institutions capable of enforcing them, while the AI race is being driven in part by the people who say its risks are real.

Daily AI news

Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.

Only what matters — every day

Follow on X