AI concerns hit a fever pitch this week when an Anthropic employee warned on X that all of humanity could be destroyed by the technology and that major frontier labs were “gambling with our lives.”
The probability of doom, or “p(doom),” is real and has been publicly discussed in the AI community for years. Elon Musk’s p(doom) has hovered between 10% and 20%, while Anthropic CEO Dario Amodei’s has been slightly higher at 10% to 25%.
But the explosive moment was reminiscent of COVID circa February 2020, when the public finally took seriously what the scientific community had been saying all along: This is no joke.
I flew to San Francisco to find the person best positioned to answer where AI is today, and whether it can be built safely ahead. On Friday, I had an exclusive hour-long sit-down with OpenAI’s Sam Altman, where we discussed the real risk AI poses to humanity, how quickly the models are developing, whether or not more powerful AI can be controlled, and what he would do if he realized AI could not be built safely. We also discussed the upcoming IPO, which he said is “ill-timed” given the safety concerns and won’t take place until 2027.
Altman told me no gamble with humanity is OK, and that he’d have no problem standing up to his investors if he needed to pause or stop AI development altogether. He also had thoughts on what’s needed from a regulatory perspective both nationally and internationally, and alluded to a pact he’s making with industry peers to slow down while safety and alignment solutions can catch up to model capabilities.
Check out the full interview with Altman above.