An OpenAI executive has warned that the world is entering a new phase of cyber threats, where AI-powered attacks are persistent rather than isolated incidents. The comments were reported by The Guardian, which spoke to the company leader about the changing nature of digital security risks.
According to The Guardian, the OpenAI figure said the industry is "hitting a different chapter" in how attacks are conducted. The warning comes alongside reporting from the BBC that OpenAI has slowed the training of its advanced AI models following a cyber-attack. The BBC’s report indicates the incident prompted a pause in development work to address security concerns.
OpenAI itself has published material on pacing model development in what it calls an era of cyber-critical capabilities, according to the linked sources. This suggests the company is weighing how quickly to push forward with new systems against the risk that those systems could be exploited or used in attacks.
The Financial Times has also weighed in, arguing that AI has not gone rogue but that the situation is worse than that, pointing to more mundane but serious risks. The Economist, meanwhile, has pushed back on fears of AI-induced armageddon, saying such concerns are overdone.
The picture emerging from these reports is one of an industry grappling with real, near-term threats rather than distant existential scenarios. The Guardian’s interview highlights that the immediate danger is not a runaway AI but persistent, targeted attacks that require constant vigilance.
What to watch next: whether OpenAI’s slower training pace becomes a broader industry trend as other labs respond to similar threats.
