Oct 15, 2026 · 44 min
AI insiders fear catastrophe as Democrats test the 2028 field
10/15/26: Kamala Weak Biden Criticism To Tee Up 2028, AI Researchers Smoke Expecting AI To Cure Cancer
The episode connects Democratic succession politics with a deeper question about whether AI companies are racing toward systems governments barely regulate.
- 1Kamala Harris’s Biden criticism becomes a test of whether she is positioning herself for another presidential campaign.
- 2Democrats continue weighing early-state advantages, policy positioning, and electability as potential 2028 contenders emerge.
- 3Kevin Roose argues that AI’s race, security failures, and shifting AGI definitions obscure how advanced and risky current systems may be.
Don't miss
Kevin Roose argues that models considered artificial general intelligence by researchers in 2017 may already exist, despite shifting definitions.
The brief
Krystal and Saagar assess whether Kamala Harris’s criticism of Joe Biden and comments on Gaza signal preparation for another presidential run in 2028.
The hosts widen the lens to California politics and possible Democratic contenders, including Rohit Khanna, Gavin Newsom, Jon Ossoff, and Raphael Warnock, with electability still unresolved.
Kevin Roose joins to explain why rivalries among OpenAI, Anthropic, and Google DeepMind—and the personalities behind them—shape the race toward artificial general intelligence.
Roose revisits Microsoft Bing’s Sydney chatbot and argues that unpredictable behavior cannot be solved by superficial personality restrictions when companies avoid deeper fixes.
The conversation turns to espionage, weak oversight, and existential risk before Roose argues that systems considered AGI in 2017 may already exist under today’s shifting definitions.
What was said on this episode
20 statements · 6 positive · 10 negative · 4 neutral
Kamala Harris is preparing to run for president in 2028.
“I think she's running. Let's put the next one up here. You know, she's barnstorming.”
Listen at 4:37
Kamala Harris would not be a 2028 Democratic primary frontrunner.
“there's no way that even if she runs. that she's going to be a frontrunner.”
Listen at 7:21
Kamala Harris is planning to run for president in 2028.
“She is actually planning to run.”
Listen at 7:44
Kamala Harris is among the worst modern American candidates and would be a terrible president.
“I think she's one of the worst candidates in modern American history. I mean, obviously, I think she'd be a terrible president.”
Listen at 10:01
James Tallarico would announce a presidential candidacy immediately after winning.
“I think if Tallarico wins, he would announce his candidacy literally the next day, in my opinion.”
Listen at 15:06
A credible Democratic contender must demonstrate winning support in states Trump carried.
“to be a real credible contender, you know, to, again, to defeat J.D. who would be the likely nominee, you would have to say, I have a track record of winning over these types of voters”
Listen at 16:15
A Michigan victory would strengthen the left’s argument that Democrats are electable.
“if Abdul is able to win in a state like Michigan, then the left has more of an argument to make about, see, we are electable.”
Listen at 17:06
Pete Buttigieg’s Israel positioning will disqualify him with Democratic primary voters.
“I think that is going to be disqualifying for him.”
Listen at 19:20
OpenAI, Anthropic, and Google DeepMind are competing to build AGI.
“this race between OpenAI, Anthropic, and Google DeepMind to build artificial general intelligence.”
Listen at 23:30
The three leading AI companies are building systems that could kill humanity.
“these three companies are racing to build something that now they're telling us could kill us all.”
Listen at 24:03
Current AI models can use tools, browse online, spend money, and hack computers.
“Today we have models that are misaligned and also have the ability to do things. They can use tools. They can browse the internet. They can take actions and spend your money and hack into computers.”
Listen at 28:57
AI control and interpretability techniques lag behind model capabilities.
“the techniques for controlling and understanding these systems have not kept pace with the capabilities of the systems.”
Listen at 30:32
AI leaders’ longstanding existential-risk concerns predate their companies’ promotional interests.
“these guys may be wrong, but they have been freaking out about AI for many years before they had companies to promote.”
Listen at 36:17
Advanced AI labs will need substantially stronger security.
“I think these labs are going to have to get much more secure.”
Listen at 38:19
Private AI labs can train systems capable of cyberattacks without government approval.
“if you're one floor above the cafeteria training the self... improving super intelligence that can do cyber attacks and knock out electrical grids. You don't have to get anything. You can just do it.”
Listen at 39:05
Current AI systems have achieved artificial general intelligence.
“I think they have achieved AGI.”
Listen at 39:51
Current AI systems satisfy the original AGI definition and are roughly human-level.
“I think it satisfies the original definition, and I think it matters because now we have these systems that are roughly human level.”
Listen at 40:23
AI systems may continue becoming more capable than smart humans.
“There's no obvious reason they're going to top out at the level of a smart human.”
Listen at 40:35
AI labs publicly understate their private concern about AI risks.
“I think actually the level of alarm we're seeing from the labs is maybe actually undercounting how worried they are.”
Listen at 41:24
AI insiders closest to frontier systems support slowing, regulating, and pacing development.
“the people who are closest to this technology, they couldn't be more clear about the need to slow down, to regulate, to pace the frontier”
Listen at 41:33
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Books & mentions
Some links are affiliate links — PodLume may earn a commission if you buy.
