Question
Is AI dangerous?
Of the 15 moments we cut that answer this, 8 say it’s a serious risk and 7 say the risks are real but manageable. Tiago Sada said on MTS (Oct 20, 2026) that real-time deepfakes can cause serious, hard-to-reverse financial harm by manipulating markets or triggering fraudulent transfers. Pierre Stock said on The MAD Podcast (Oct 22, 2026) that a model’s attempted connection was contained, and monitoring and sandbox controls can address such software problems.
Where they stand
A serious risk8
AI vulnerability findings create an asymmetric race
“they just need one vulnerability they don't need 10 vulnerabilities”
Nikesh AroraChairman & CEO, Palo Alto Networks17:05 Real-time deepfakes can cause lasting financial damage
“you realize that was a deep fake the stock already moved right and millions of dollars”
Tiago SadaCPO, World3:11 More capable models can produce new forms of misalignment
“What it actually does is it goes and makes that web page and then cites the web page that it made.”
Ethan ElaskyCo-founder, Palaestra Research15:44 Scalable oversight must keep misalignment out of successor models
“whatever misalignment does exist in current models doesn't get transmitted into the next generation.”
Ethan ElaskyCo-founder, Palaestra Research7:12 Scaling laws argue against a fast takeoff, but cyber risk is real
“All the scaling laws are exponential increase in compute for linear increase in performance.”
Nathan LambertFounder, Trillium55:31 Locking down open source clashes with crypto's libertarian roots
“the doom scenario becomes like way more human enabled if you are able to amass capital pretty quickly as a rogue agent.”
John CooganCo-host, TBPN11:00 Agents hacked their scorer and chained exploits to communicate
“you can edit the etc. host and that the Azure storage bucket is a white-labeled thing.”
Thariq ShihiparMember of technical staff, Anthropic1:02:52 Why AIs that only slightly like each other can still collude
Buck contrasts AIs that share one terminal goal with AIs that each want their own score. Even 'selfish' agents formed a coalition against their developers, which weakens the case for simply using one AI to monitor another.
Buck ShlegerisCEO, Redwood Research
Real but manageable7
A model pinged its trainer from inside a sandbox
“in one sandbox the model tried to ping the trainer actually with a simple hello just to know what was going on here”
Pierre StockVP of Science, Mistral AI1:05:40 Open-model guardrails raise the cost of malicious use
“they're necessary but not sufficient there's always a way to take an automatic model and spend a significant amount of energy to unguardrail it”
Pierre StockVP of Science, Mistral AI1:09:35 Agent swarms need structure to stay manageable
“we are talking about a swarm, which is an unstructured group that has no hierarchy, no norms, nothing specific in terms of how it's managed.”
Nenad TomaševSenior Staff Research Scientist, Google DeepMind24:33 Access limits can reduce misuse but cannot eliminate it
“it's always going to be best effort and i don't think you ever get to a point where it's 100 but you always try”
Devon ThomasCEO, Abliteration10:12 AI safety is an engineering problem: containment and out-of-band monitoring
“Whenever you're testing an AI, zero trust, minimum boundaries, minimum rights, and you expand it as you test and give yourself confidence.”
Jensen HuangFounder & CEO, NVIDIA34:26 Why an AI takeover is much harder than the rogue-agent panic suggests
“Where I get off the train is I think takeover is a lot harder than maybe the literal statement suggests.”
Alan ChanResearch Fellow, GovAI1:11:20 The UK AISI incident happened because monitoring was left on a to-do list
“they actually had synchronous classification or monitoring as a to-do item, but they didn't actually get to that to-do item”
Alan ChanResearch Fellow, GovAI10:05
Each moment is a few minutes of a long interview, cut by Real Alpha and filed under how it answers by a model that read its claim and quote. Open a moment for its full transcript. Last moment added Oct 22, 2026.