Sam Altman says some AI accidents are 'unavoidable' as the industry grapples with apocalypse fears

Sam Altman says immoderate AI accidents are unavoidable, but it's really you study from them.

Speaking astatine Salesforce's Dreamforce convention connected Tuesday, the OpenAI CEO said tech companies should clasp a civilization of transparent mishap reporting, akin to that of the aviation industry.

"I do deliberation that immoderate grade of accidents are unavoidable pinch caller technology," Altman said. "What I really attraction astir is that we person sewage a awesome civilization of mishap reporting and learning."

He cited the Federal Aviation Administration and the National Transportation Safety Board arsenic examples the AI manufacture should follow, saying the agencies' accent connected investigating incidents has helped make flying safer.

"You know, accidents are going to happen," he said of the FAA's and NTSB's approach. "We're going to study arsenic overmuch arsenic we tin and respond from each one."

Altman's comments travel arsenic the AI manufacture faces renewed questions astir whether its biggest labs are moving excessively quickly toward powerful autonomous systems.

Jacob Coxon, a interrogator who discontinue Anthropic past week and antecedently worked astatine OpenAI, kickstarted a statement by saying that the companies were "racing consecutive to self-improving superintelligence and gambling pinch our lives."

'The world is correct to beryllium acrophobic of this'

Other AI information researchers person since told Business Insider that precocious systems could 1 time evade quality power aliases make cyber, biological, and different catastrophic threats easier to transportation retired — although they would look hurdles successful accessing the beingness world.

Altman acknowledged that AI last day fears are legitimate. "The world is correct to beryllium acrophobic of this," he said astatine Dreamforce, pointing to the anticipation of a loss-of-control mishap and the consequence that a mini number of AI developers could ore excessively overmuch power.

He said that the reply is not to wantonness the technology, and that AI companies must alternatively guarantee that "alignment and information and monitoring" stay "way up of capabilities." He added that they should besides beryllium prepared "to slow down aliases stop" if they cannot do so.

His comments whitethorn besides sharpen a cardinal disapproval from AI-safety advocates: that frontier labs opportunity they admit the stakes yet proceed building much tin models.

Altman said "people get very scared" erstwhile it seems for illustration commercialized pressures mightiness get successful the measurement of companies aliases countries doing "the correct thing."

The constituent is not to judge disaster, Altman said, but to disclose incidents, study from them quickly, and forestall them from escalating.

"Maybe we tin avoid, you know, ever getting to those kinds of stakes," he said.

The statement is besides unfolding arsenic the Trump management has shown small liking successful regulating the sector, seeking to support the US's lead complete China.

David Sacks, a statesmanlike advisor connected subject and tech and the erstwhile White House AI Czar, reiterated that stance connected Sunday, responding to calls from OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei to "pace the frontier."

"Dario has written that we request to 'pace the frontier,' and Sam has agreed. People whitethorn beryllium amazed by my response: spell ahead," Sacks wrote connected X. "But extremity pretending you request anyone else's permission."