OpenAI researcher warns ultrafast AI could leave security teams in the dust

An OpenAI researcher who goes by the pseudonym "roon" warns that extremely fast AI inference creates security risks that current safeguards can't handle. A misaligned model running at the level of today's best systems but 50 times faster could infiltrate systems so quickly that human response teams wouldn't stand a chance, roon says. Monitoring alone wouldn't cut it. "You need autonomous detection and shutdown, not just monitoring," roon writes.

The warning taps into a growing concern in AI security research as models get more capable. The potential for harm grows when models aren't properly aligned with human goals, and that alignment problem remains unsolved. Speed makes it worse. Roon's point: if attacks get automated, defense has to follow suit.

Roon's comment came in response to OpenAI's unveiling of its new AI chip, which can outperform current hardware in inference speed. OpenAI and Anthropic already offer "Fast Modes" that give paying users access to quicker AI models.

AI News Without the Hype – Curated by Humans

Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.