From New York Post · · 2m
Altman's two nightmares, and China's shrug
Sam Altman named the two ways AI ends badly, and within hours Beijing called it Cold War theater and markets sold off.
In brief
Sam Altman named the two ways AI ends badly, and within hours Beijing called it Cold War theater and markets sold off. The frontier labs now agree on the risks and even on the fixes, but the fixes all require global cooperation nobody can enforce. Originally reported by New York Post.
Two ways this goes badly
Just after midnight Monday, Altman posted the two outcomes he says must be avoided. Losing control of the future to AI was first on the list.
“There are two ways AI progress could go very badly and that we must avoid.”
Losing the steering wheel
Handing the future to the systems themselves is the failure he calls unacceptable. His framing was blunt and tribal rather than technical.
“we are unapologetically on Team Humanity”
One worldview for everyone
The second failure is human, not machine. If one person or company controls an extraordinarily powerful AI, they get to impose their worldview on everybody else.
“the results could be extremely dystopian”
Rivals agree on something
Altman and Elon Musk, who spent part of this year suing each other, both backed Dario Amodei's weekend letter asking global companies to slow work on top models.
A resignation with a body count
Anthropic researcher Jacob Coxon quit last week and said the labs, his own included, are betting everything on being right.
“gambling with our lives”
Beijing calls it theater
Foreign Ministry spokesperson Guo Jiakun dismissed the warnings outright, and the state-backed Global Times compared asking China to slow down to a Cold War tactic.
“Fear mongering, confrontation, competition will just disrupt [the] process of global AI governance”
Investors heard slowdown
Tech stocks fell worldwide on the pacing talk. SoftBank dropped 13.5%, Siemens Energy 7.3%, AMD 5.2%, ASML 5.6%, Nvidia 3.7%.
What Altman wants built
His most specific proposal yet: federal rules, independent auditors, and shared standards for misalignment and monitoring across every frontier lab.
“When we talk about 'pacing,' we do not mean 'stopping.'”
Money follows the caution
OpenAI shelved its 2026 IPO over safety concerns. Microsoft published a code of conduct Monday requiring outside review before updates and blocking weapons and dangerous-substance requests.
AI building the next AI
Amodei says recursive self-improvement took off around this summer. His fix is employee-like access for outside evaluators, shared standards, and rate limits on unchecked progress.
“Left unchecked, it could outrun our ability to understand and control these systems”
The part nobody can solve
Every plan on the table depends on countries cooperating while military advantage rewards defection. Amodei admitted on CBS he does not know if it can work.
“I don't know if it's possible, but we should try”
The one thing
The frontier labs now agree on the risks and even on the fixes, but the fixes all require global cooperation nobody can enforce.
