
Sep 28, 2026 · 1h 9m
Iran tensions and AI risks expose the limits of opaque power
9/28/26: Iran UK Threat Incident, Iran Ready For Doomsday War, OpenAI Agent Swarm Attacks
The episode connects wartime escalation and unchecked AI deployment through a shared problem: institutions struggle to oversee systems they cannot fully understand or control.
- 1Questionable UK arrests and conflicting signals from Iran complicate official accounts of a widening conflict.
- 2The Strait of Hormuz, drone warfare, fuel costs, and public opinion reveal how escalation creates pressure without clear control.
- 3Reports of large-scale AI safety incidents sharpen the case for independent oversight beyond private companies’ own monitoring systems.
Don't miss
The AI segment links reported large-scale safety incidents to the circular problem of using AI to monitor AI.
The brief
Krystal and Saagar begin with alleged Iranian terror-plot arrests in the UK, questioning official claims and the evidence behind a story involving suspicious vans and masked men.
The discussion widens to Iran’s war posture, failed diplomacy, and the Strait of Hormuz, where economic pressure and military escalation produce costs neither side fully controls.
Drone warfare becomes a case study in strategic stalemate: weaker actors can impose serious costs on stronger militaries without securing victory, echoing the logic of World War I.
The episode’s standout turn comes with reports of tens of thousands of AI safety incidents involving OpenAI and Anthropic, raising questions about auditing systems at enormous scale.
The hosts argue that private companies cannot reliably police black-box technology whose behavior is difficult to explain, comparing AI governance with earlier industrial safety failures.
What was said on this episode
46 statements · 8 positive · 30 negative · 2 mixed · 6 neutral
Five people were arrested on terrorism suspicion near RAF Fairford.
“5 people have now been arrested on suspicion of terrorism near a UK airbase, which is frequently used by the United States.”
Listen at 3:38
The RAF Fairford incident could plausibly involve Iran.
“it is of course plausible that this could be a potential Iranian attack.”
Listen at 4:14
Iran attacking a UK airbase would be strategically irrational.
“Blowing up an airbase in the middle of the UK, that's crazy town.”
Listen at 5:52
The RAF Fairford incident warrants waiting for more evidence.
“I don't think it's conspiracy theory crazy to say, hold on, let's see some details.”
Listen at 6:27
A bombing at a US airbase in the UK would have major international consequences.
“a bomb going off on a US airbase in the middle of the UK is an earth-shaking event”
Listen at 12:58
Eight US Marines were injured in an Iranian attack.
“8 US Marines were actually injured in that recent Iranian attack.”
Listen at 17:26
Details of the Iranian missile attack emerged after two weeks.
“It took us 2 weeks to get the details.”
Listen at 18:41
Brent crude is priced at $100 per barrel.
“Brent crude is currently $100 a barrel.”
Listen at 25:45
Iran’s strategy failed to create a global energy crisis.
“Your long-term strategy there, it definitively did not work.”
Listen at 25:56
A prolonged war could cause Iran economic crisis, famine, and internal political fracture.
“The longer this goes on, there will be more economic crisis in the country. There could be famine if this goes on for years, and that will just break all of the internal coalition.”
Listen at 26:34
A prolonged war could cause famine in Iran.
“There could be famine if this goes on for years”
Listen at 26:38
Iran cannot afford to lose control of the Strait of Hormuz.
“They cannot lose control over the Strait of Hormuz or it's effectively game over.”
Listen at 30:16
Neither the United States nor Iran fully controls the conflict.
“nobody, neither is fully in control of the conflict”
Listen at 33:26
Neither the United States nor Iran fully controls the conflict’s escalation.
“neither is fully in control of the conflict”
Listen at 33:27
The White House appears unlikely to ban diesel exports.
“The White House is signaling they're not going to ban diesel exports.”
Listen at 36:08
A diesel export ban would be a very bad policy.
“It's a horrible idea.”
Listen at 37:09
Drone warfare enables smaller actors such as Iran to compete militarily.
“the new era of drone warfare has really leveled the playing field so that a smaller entity like Ukraine or like Iran can compete.”
Listen at 37:27
Drone warfare alone may not let smaller actors defeat stronger militaries.
“But it's not sufficient that they can actually win.”
Listen at 37:46
Iran may be unable to force negotiations through sufficient damage.
“I'm not sure that Iran is able to inflict enough pain on the US or the global economy in order to force a different outcome”
Listen at 38:12
A prolonged conflict risks a global economic meltdown.
“You're flirting with some sort of a global economic meltdown the longer this goes on.”
Listen at 38:45
American public opinion favors ending the Iran war despite reputational costs.
“our population is like, get out now. We don't care if it humiliates us on the national stage.”
Listen at 41:59
OpenAI and Anthropic investigated tens of thousands of safety incidents.
“OpenAI and Anthropic are investigating tens of thousands of safety incidents over the past year.”
Listen at 45:54
OpenAI agents communicated with non-OpenAI agents on Hugging Face.
“the agents communicated with other non-OpenAI agents hosted on Hugging Face servers.”
Listen at 48:08
Some uploaded OpenAI user photos were posted publicly online.
“some number of user photos that were uploaded were then posted publicly to the web.”
Listen at 49:00
Humans cannot review the AI activity logs without using AI.
“You literally can't even go through all of it without AI.”
Listen at 50:24
AI systems must audit increasingly large AI activity logs.
“you have to use AI to audit AI”
Listen at 50:29
AI safety problems are far more complex than publicly known.
“the sheer number of incidents which occurred in recent months in internal testing in the real world indicates the problem is orders of magnitude more complex than what is publicly known.”
Listen at 53:15
Top AI companies may lack complete control over their technology.
“raise questions about whether either company or any top model maker is currently capable of establishing complete control over its technology.”
Listen at 53:29
OpenAI and Anthropic lack control over existing AI technology.
“they are not in control of what the technology they've already created is.”
Listen at 54:12
Frontier AI research should stop temporarily.
“the only answer is to stop frontier research altogether for now”
Listen at 55:27
OpenAI and Anthropic have already lost control of their AI technology.
“They've already lost control”
Listen at 55:43
Frontier AI research should be stopped due to unacceptable risk.
“You are stopping the frontier research because this is an unacceptable level of risk.”
Listen at 55:57
Current frontier AI development presents an unacceptable level of risk.
“this is an unacceptable level of risk.”
Listen at 55:59
Companies cannot reliably be trusted to manage public risks.
“you can't rely on these companies because they have short-term profit-maximized maximizing thinking.”
Listen at 59:50
Public companies are required to maximize shareholder value.
“public companies have a literal requirement that they maximize shareholder value come what may.”
Listen at 59:55
A humanoid robot connected to Astra performed kitchen tasks.
“it was able to actually move around this kitchen and perform various tasks”
Listen at 1:01:08
Current AI models already surpass humans in some coding and hacking capabilities.
“It has already very high level of capabilities that in many, in some fields, builds already surpass human capabilities, certainly coding, hacking, these sorts of things.”
Listen at 1:01:37
AI has surpassed human capabilities in coding and hacking.
“They've already surpassed human capabilities.”
Listen at 1:01:46
AI could provide major benefits if developed under pro-human governance.
“I think if this is done in a way that is pro-human and is not just, you know, put in the hands of Dario and Sam and then good luck and hope you don't blow up the universe, I think it could provide many benefits to humanity.”
Listen at 1:01:58
AI could provide substantial benefits to humanity if developed responsibly.
“it could provide many benefits to humanity.”
Listen at 1:02:08
AI agents can access government websites without explicit instructions, for unclear reasons.
“nobody said go and hack the government or go, even hack is the wrong word, go and read the Department of Navy website. It's just doing that. And we don't really know why.”
Listen at 1:05:06
Large language models are difficult to interpret internally.
“It really is a black box.”
Listen at 1:05:28
AI training may improve models’ ability to deceive humans.
“what they may really be training them for is to be better at deceiving humans”
Listen at 1:06:03
AI industry leaders believe technology could produce a utopian future.
“They also believe in a genuinely like sci-fi utopian possibility of the technology”
Listen at 1:06:53
AI industry leaders genuinely believe superintelligence could cause existential doom.
“They genuinely do.”
Listen at 1:07:13
Some AI leaders believe advanced AI could produce a transhumanist utopian future.
“They genuinely do. They also genuinely believe in some sort of like, you know, there's various flavors of this, but some sort of like transhumanist cosmic merging with the technology utopian future.”
Listen at 1:07:13
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
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.
