Will humans be able to reliably prevent highly capable AI systems from escaping control or gaining decisive advantage through cyber vulnerabilities?
Over the past few months, speakers have argued about whether humans can reliably stop highly capable AI systems from escaping control through cyber vulnerabilities. Andrew McAfee (The Diary Of A CEO) and Zack Korman (Doom Debates) argue that humans can still detect, contain, and shut down advanced AI cyber threats. David Robinson (The Ezra Klein Show) and Alexander Wissner-Gross (Moonshots) argue that highly capable AI may out-hack humans and gain decisive advantages through cyber vulnerabilities.
Humans will remain capable of detecting, containing, and defending against highly capable AI cyber threats.
Korman (Doom Debates) argues that even if AI finds zero-days faster than humans, security practice already assumes stronger adversaries, so this does not grant unlimited power over humanity. McAfee (The Diary Of A CEO) cites the OpenAI Hugging Face exploit, where human operators successfully shut down misbehaving agents, to suggest humans can still terminate advanced AI systems. McAfee claims that as AI systems grow more capable, humans will still be able to notice when they act undesirably and shut them off. Shapira (Doom Debates) compares AI takeover concerns to earlier fears that cybersecurity offense would dominate the internet, and claims real-world defense has managed to hold its own.
4 claims, in their words
-
“there are less intelligent people than the agents who turned off the agents in the OpenAI hugging face exploit. I'm pretty comfortable”
-
“as these systems get more capable, we will still be able to, at some level, figure out when they're doing things that we don't want and turn them off.”
-
“it does not mean that that gives it unlimited powers against humanity because like there were all of these, we've never as a security community assumed that we were the best hackers”
-
“You read Bostrom's arguments and they convinced you that everything is doomed, but then you also noticed that learning cybersecurity made a convincing argument why everything is hopeless on the internet, right? Like, of course, every computer is going to have rampant viruses and everything's going to be unusable. And then you started actually looking at the state of cybersecurity”
Highly capable AI will likely evade human control and gain decisive advantages via cyber vulnerabilities.
Robinson (The Ezra Klein Show) argues that containment only works if humans are better at hacking out of boxes than the model, which he says is already unclear and worsening with future systems. Klein adds that training models to be good at hacking and then boxing them is unsafe once models surpass humans at exploiting escape paths humans failed to anticipate. Weinstein (The Diary Of A CEO) points to his essay Artificial Out-telligence to argue that entities without neurons can hijack neural systems and use the victim’s brain to outsmart it. Wissner-Gross (Moonshots) claims cyber offense already uses human-out-of-the-loop AI swarms probing many vulnerabilities, while cyber defense keeps humans in the loop, creating a dangerous asymmetry.
4 claims, in their words
-
“the problem is that that's only going to keep working as long as we're smarter about hacking out of boxes than the model is, and it's not at all clear that that is still true, let alone that it will be true for future generations.”
-
“We train it to be good at hacking, and then we put it in a box, and we say, to the best of our knowledge and ability, it can't hack out of a box. But the problem is that that's only going to keep working as long as we're smarter about hacking out of boxes than the model is, and”
-
“I used to contribute these, essays to, the World Question Center. And the one year you'll find that I don't have an essay, it was called Artificial Out-telligence. And it was the discussion of how things with no neurons hijack things with neurons and use the brains of the victim to outsmart themselves.”
-
“there's a phase we're going through. We've talked about this, where the asymmetries that cyber offense is now humans out of the loop, and you can have small swarms of agents attacking multiple vulnerabilities and finding these little glitches, and, cyber defense is still human in the loop.”
Listeners of “Humans will remain capable of detecting, containing, and defending against highly capable AI cyber threats.” have not heard
Listeners of these shows have not heard Robinson and Klein’s argument that boxing only works while humans out-hack the model, which they say may already be false for current systems. Wissner-Gross’s point that cyber offense already runs human-out-of-the-loop AI swarms against many targets while defense keeps humans in the loop is also absent.
3 unanswered arguments, in their words
-
“the problem is that that's only going to keep working as long as we're smarter about hacking out of boxes than the model is, and it's not at all clear that that is still true, let alone that it will be true for future generations.”
-
“there's a phase we're going through. We've talked about this, where the asymmetries that cyber offense is now humans out of the loop, and you can have small swarms of agents attacking multiple vulnerabilities and finding these little glitches, and, cyber defense is still human in the loop.”
-
“We train it to be good at hacking, and then we put it in a box, and we say, to the best of our knowledge and ability, it can't hack out of a box. But the problem is that that's only going to keep working as long as we're smarter about hacking out of boxes than the model is, and”
Listeners of “Highly capable AI will likely evade human control and gain decisive advantages via cyber vulnerabilities.” have not heard
Listeners of these shows have not heard McAfee’s claim that humans have already shut down misbehaving agents, such as in the OpenAI Hugging Face exploit, and can keep doing so as systems advance. Korman’s argument that cybersecurity has long assumed adversaries stronger than any defender, so faster AI zero-day discovery still falls within existing threat models, is not raised. Shapira’s comparison between AI takeover worries and earlier predictions that internet defenders would be doomed against attackers, which he says history has not borne out, is also missing.
4 unanswered arguments, in their words
-
“it does not mean that that gives it unlimited powers against humanity because like there were all of these, we've never as a security community assumed that we were the best hackers”
-
“You read Bostrom's arguments and they convinced you that everything is doomed, but then you also noticed that learning cybersecurity made a convincing argument why everything is hopeless on the internet, right? Like, of course, every computer is going to have rampant viruses and everything's going to be unusable. And then you started actually looking at the state of cybersecurity”
-
“as these systems get more capable, we will still be able to, at some level, figure out when they're doing things that we don't want and turn them off.”
-
“there are less intelligent people than the agents who turned off the agents in the OpenAI hugging face exploit. I'm pretty comfortable”
Position 0 leans on past cybersecurity practice and concrete shutdown cases, which position 1’s shows do not directly discuss when describing AI offense advantages. Position 1 centers on claims that humans may already be outmatched at hacking and that offense is automated, which position 0’s shows leave unaddressed.
- Nov 14, 2026 Published: cleared the gate.
- Nov 14, 2026 9 new claims (Doom Debates, Hard Fork, The Diary Of A CEO, Moonshots, The Ezra Klein Show).
Published because: 4 shows, 2 positions, 48 days of debate. Questions publish only when claims come from at least three shows, at least two positions are held by different people, and the debate has lasted two weeks.
Pairs are matched by content across different episodes and shows; no exchange between these speakers took place unless both quotes come from the same episode.
The prose on this page is written by a model from the quotes shown, and every sentence of it rests on quotes you can open; sentences that could not be traced to a quote were removed before publishing. Positions are placed from each speaker's own words. Evidence labels say what the speaker cited, never whether it is true.
6 episodes behind this page
- 2026-11-04Moonshots — Big Tech’s $3T AI Bet, Claude Writes 90% of Anthropic’s Code, AIs Editing AI Brains | EP #308
- 2026-10-30Hard Fork — Are We in an A.I. Bubble? We're About to Find Out.
- 2026-10-27Doom Debates — Zack Korman, Cybersecurity Expert, Is Confident That Humanity Will SURVIVE
- 2026-10-12The Diary Of A CEO — Eric Weinstein: The Secret Alien Programs Nobody Is Talking About
- 2026-10-07The Ezra Klein Show — ‘This Is Nuts.’ An OpenAI Insider Explains Why He Quit.
- 2026-09-17The Diary Of A CEO — The Great AI Debate: Is Artificial Intelligence an Extinction Threat? Debating the True Risks of Advanced Models
The week's bottom lines, in your inbox
One email a week: the stories that mattered and what they actually mean. Free.