Breaking

OpenAI's Altman Says AI Firms Can Develop Safely as Frontier Risk Debate Widens

September 15, 2026 at 20:31 EDT

AI Agents
Policy & Regulation
Foundation Models

OpenAI CEO Sam Altman said AI companies can continue developing more powerful systems safely, stepping into a fast-moving safety debate at Salesforce's Dreamforce conference. His remarks came days after Anthropic CEO Dario Amodei urged frontier labs to adopt a slower frontier pace while monitoring, alignment and outside checks catch up.

Sep. 15, 2026 · OpenAI

Altman says safety can scale with AI

The OpenAI CEO argued frontier labs can keep advancing if they report incidents, strengthen controls, and slow when safeguards cannot keep up.

6-12 months

warning window raised in the wider safety debate

3 steps

embedded evaluators, standards, global coordination

93%

message-board tasks came from unsolved eval items

198/898

ExploitGym tasks had no prior model solution

Warning window, drawn to scale

The debate centers on how quickly stronger agent swarms could arrive if controls fall behind.

6 months

near end

12 months

far end

How pacing is supposed to work

Capability gains

→

Incident reporting

→

Embedded review

→

Slow if controls lag

Confidence case

Frontier labs say transparent incident reporting, stronger sandboxes and outside evaluators can let development continue responsibly.

Critical case

Skeptics worry voluntary pacing could be too weak, too slow, or tilted toward large closed labs that can absorb compliance costs.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.