
Sam Bankman-Fried, the disgraced FTX founder now serving a 25-year federal prison sentence for fraud, estimated that artificial intelligence has roughly a 25% chance of destroying humanity. He shared that figure during a phone interview from prison in Lompoc, California, with Eli Goldfine, a 14-year-old prediction-markets podcaster.
The estimate, which Bankman-Fried said depended partly on whether humans and AI end up coexisting peacefully or AI ultimately wipes out humanity, was reported by Business Insider. According to the outlet, Bankman-Fried also offered a wider range, saying the probability of an AI-driven catastrophe could fall anywhere between 20% and 60% depending on gray areas in how one defines catastrophe. He said estimates vary widely depending on what people consider a genuinely catastrophic event.
The conversation took place over a series of phone calls on Saturday and Sunday that lasted roughly 90 minutes in total, per the outlet's reporting. The calls were recorded and monitored, as is standard for federal inmate phone lines, and covered more than just AI — the pair also talked through FTX's downfall and Bankman-Fried's interest in effective altruism.
Who Is the Teenager Behind the Interview
Eli Goldfine, who is based in Westchester County, New York, publishes a newsletter and podcast called Supercycle featuring interviews with industry figures, the report notes. His conversation with Bankman-Fried placed him in rare company: few reporters have gotten extended phone access to the former billionaire since his 2023 conviction.
A jury found Bankman-Fried guilty of fraud and money laundering in 2023 after prosecutors detailed a scheme that involved $11 billion in commingled FTX customer funds and Alameda Research funds. He is now 34 years old and, according to Business Insider's reporting, scheduled to remain in prison until 2043. In September, he asked the U.S. Supreme Court to consider his appeal.
The Anthropic Stake That Got Away
Part of the interview touched on Bankman-Fried's history with Anthropic, the AI company he once helped fund. He was the leading investor in Anthropic's April 2022 Series B round, putting in $500 million for a stake that was worth about 13.6% of the company at the time, per the outlet's account. His shares were sold off in 2024 for $1.3 billion, with the proceeds used to help pay FTX creditors, including fraud victims.
Had he held onto that stake, the numbers now look almost unfathomable. Anthropic shares traded at around $97 after the company's initial public offering last Thursday, a price at which Bankman-Fried's original holding would have been worth more than $40 billion, the report states. He told Goldfine he remains bullish on Anthropic despite having been forced to part with his stake.
Brain Uploads and 'Synergistic' AI
Beyond the probability estimates, Bankman-Fried described a scenario in which AI-driven work guided by human goals could produce massive technological expansion, framing it as a more optimistic counterpoint to his extinction-risk numbers. He also told Goldfine he might consider uploading his brain to the internet, according to the interview as reported by Business Insider.
The term p(doom), used in AI-safety circles to mean the probability of an AI-driven catastrophe that wipes out humanity, frames how researchers and commentators like Bankman-Fried quantify these risks. His 25% figure sits well above what the field's own experts have reported. A 2023 survey of 2,778 AI researchers completed in October of that year, run by AI Impacts, found a median estimate of just 5% for AI causing human extinction or similarly severe, permanent disempowerment, though the mean estimate was considerably higher at 16.2%. The same survey found one in ten respondents put the odds of such extremely bad outcomes at 25% or higher, close to Bankman-Fried's own number.
That same research, detailed on the AI Impacts Wiki, also found the aggregate forecast among researchers gave a 50% chance of high-level machine intelligence arriving by 2047 — thirteen years earlier than the same survey estimated just a year prior. Seventy percent of those surveyed said AI safety should be prioritized more, or much more, than it currently is.









