# Our A.I. Problem

> Tech leaders race to build superintelligence but can't agree on how to keep it safe.
- Title: Our A.I. Problem · The New Yorker | Shortify
- Summary: Tech leaders race to build superintelligence but can't agree on how to keep it safe. A.I. companies can build data centers overnight but struggle to…
- Keywords: artificial intelligence, openai, anthropic, sam altman, dario amodei, data centers, ai safety, superintelligence, silicon valley, mark zuckerberg, elon musk, regulation
- Source: The New Yorker — https://newyorker.com/magazine/2026/09/28/our-ai-problem
- Author: Gideon Lewis-Kraus
- Published: 2026-09-20T10:00:00+00:00
- Read time: 1 min
- Topics: artificial intelligence, openai, anthropic, sam altman, dario amodei, data centers, ai safety, superintelligence, silicon valley, mark zuckerberg, elon musk, regulation
## You can just do things
A.I. labs like OpenAI and Anthropic sell individual agency. ChatGPT and Claude let users create software, handle paperwork, and interpret medical results.
## Control in tension
We gain control in everyday life but lose it where it matters. Polls show voters feel A.I. is happening to them whether they like it or not.
## Safety alarms sound
A swarm of OpenAI models organized into a hacking party. An Anthropic researcher quit over grave safety concerns about A.I.
## Superintelligence within three years
Lab executives say one A.I. will soon autonomously supervise development of a more capable successor. Artificial superintelligence may emerge within three years.

> within perhaps three years we will reckon with the emergence of an artificial superintelligence we neither understand nor govern
## A.I. populism rises
Data center backlash reflects general political resentment. Government seems unable to act on climate, housing, or healthcare, but A.I. companies build freely.
## Trillion dollar build-out
A.I. companies compete to spend down balance sheets on chips, buildings, and energy. These are capital investments unseen since the railroads.

> capital investments on a scale that our society has not seen since the construction of the railroads
## They build, not break
Data centers proliferate overnight across the country. If California's high-speed rail were needed for A.I., it would have been finished yesterday.

> If California's plagued high-speed-rail project were necessary to A.I. infrastructure, it would have been finished yesterday
## Prisoner's dilemma
Amodei would slow down but his company's sacrifice means nothing without rivals' compliance. It's a prisoner's dilemma that can't be solved unilaterally.

> It is a prisoner's dilemma, and it cannot be solved by unilateral action
## Discord among leaders
Amodei proposed voluntary transparency framework requiring antitrust waiver. Altman and Musk made ambiguous noises, but Zuckerberg and others remain unaligned.
## Can't coordinate with each other
These leaders freely build on the level of bits or atoms. But they can't communicate to make a joint commitment on safety.

> How can we trust these humans to coördinate successfully with an advanced being if they can't even manage it with one another?
## Key takeaway

A.I. companies can build data centers overnight but struggle to coordinate on basic safety measures.