# Altman names two ways AI could go very badly

> After a resignation warning and agentic AI security incidents, the heads of OpenAI and Anthropic are suddenly talking about hitting the brakes.
- Title: Altman names two ways AI could go very badly · PCMAG
- Summary: After a resignation warning and agentic AI security incidents, the heads of OpenAI and Anthropic are suddenly talking about hitting the brakes. The biggest…
- Keywords: ai safety, openai, anthropic, regulation, superintelligence, technology, Altman, names, two, ways, very, badly
- Source: PCMAG — https://pcmag.com/news/openai-ceo-warns-of-2-ways-ai-could-go-very-badly-if-tech-doesnt-slow-down
- Author: James Peckham
- Published: 2026-09-14T15:11:47+00:00
- Read time: 1 min
- Topics: ai safety, openai, anthropic, regulation, superintelligence, technology
## AI bosses talk brakes
Over one weekend, the leaders of the top labs posted about slowing large language model work, after security incidents involving agentic AI and a resignation warning that went viral.
## Altman joins the chorus
Sam Altman first agreed with Anthropic's Dario Amodei in a short Saturday tweet, then posted a longer message warning of ways "AI progress could go very badly."

> AI progress could go very badly
## The resignation that lit it
An Anthropic researcher quit last week over safety, saying OpenAI and Anthropic are both gambling with our lives by pushing self-improving AI.

> gambling with our lives
## Risk one: losing control
Altman called handing the future to AI unacceptable, though he never spelled out what losing control would look like for OpenAI.

> We are unapologetically on Team Humanity, and AI must always serve people.
## Risk two: power piles up
His second worry is concentration: one country or one lab pulling far ahead, or a single owner using powerful AI to push their worldview on everyone.

> the results could be extremely dystopian
## Rules without waiting
Altman says OpenAI would welcome a federal framework of consistent safety requirements, but that shared standards for misalignment, monitoring and safety could start before Congress acts. He offered no specifics.
## Amodei's three steps
Anthropic's plan: third-party evaluators inside each company, one set of standards across firms, and agreements with foreign governments to block dangerous uses.
## Just slow down yourselves
White House science adviser David Sacks said labs should pause on their own instead of demanding cover, accusing them of dressing up a cartel as regulation.

> Stop pretending you need anyone else's permission.
## Washington isn't buying it
Trump posted that a high-IQ president is the only guardrail AI needs and jabbed at Amodei. JD Vance called the industry's plea for regulation a possible Trojan horse.

> It feels a little bit to me like a bit of a Trojan horse.
## Key takeaway

The biggest AI labs are now publicly asking for rules they could simply adopt themselves, and Washington is calling out the gap between the ask and the action.