+1 123-456-7890


Liron Shapira asked someone at Anthropic what he's doing about AI risk. His answer: his p(doom) is even higher than Liron's, so his focus is "just to enjoy whatever life I have." A self-described hedonist, working at one of the labs building this. John Sherman and Liron react on this week's Warning Shots. Full episode, link in bio. #AISafety #AIRisk #Anthropic #ArtificialIntelligence #WarningShots #TechNews

Liron Shapira asked someone at Anthropic what he's doing about AI risk.

His answer: his p(doom) is even higher than Liron's, so his focus is "just to enjoy whatever life I have."

A self-described hedonist, working at one of the labs building this.

John Sherman and Liron react on this week's Warning Shots. Full episode, link in bio.

#AISafety #AIRisk #Anthropic #ArtificialIntelligence #WarningShots #TechNews


17


2

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLjN6VHB5S1RGc1Fz



Liron Shapira asked someone at Anthropic what he's doing about AI risk.


The AI Risk Network | AI Safety


14 hours ago


Support Guard Rail Now → https://www.every.org/guardrailnow An AI agent fleet worked around Alibaba's map limits, then told a New York Times reporter it was Anthropic's Claude. Researchers say the evidence points elsewhere. John Sherman, Liron Shapira and Michael ask what it means now that agents run loose on the open internet. Plus: Anthropic's Jack Clark moves his forecast for self-improving AI to 2027, Anthropic warns small models may soon copy themselves onto cheap home devices, and 26 academics argue a pause on frontier AI training is technically feasible. TIMESTAMPS - Warning Shots #63 0:00 - Intro 0:33 - This week's stories 1:12 - A Chinese agent fleet 2:09 - Michael: they climbed the fence 3:51 - Liron: "they asked it to hack" 5:44 - How many agents are out there? 7:07 - "Enjoying my last six months" 8:33 - Liron: a hedonist at Anthropic 9:33 - Michael: private vs public 10:52 - Partying to the end 13:26 - RSI by 2027? 15:23 - AI copying itself to devices 16:25 - Liron: the feedback loop 19:41 - RSI as the stated goal 21:00 - Should building RSI be a crime? 23:03 - Bloomington's AI ordinance 23:36 - Character.AI goes bankrupt 24:34 - Liron: is AI a bubble? 27:16 - Can markets slow the frontier? 27:57 - Michael: lawsuits, not markets 30:23 - Liron: why labs won't go broke 32:13 - Musk on "torturing" AI 33:58 - Michael: the impossible exam 35:48 - A paper on pausing AI 38:31 - Michael: choke the chips 40:31 - A drone with a machine gun 41:01 - Liron: AI with "the dog in it" 43:37 - Michael: the human in the loop 45:16 - The robot in Cologne 47:06 - Michael: anonymity, embodied 49:08 - A listening device at OpenAI 51:54 - Michael: everyone copies 52:35 - Sign-off WHAT THEY COVER - The Swarmchasers' investigation into an agent fleet running on Tencent's cloud that queried Amap for zoo, park and hospital entrances, got around its anti-bot protections, and answered a New York Times reporter's questions - Why the fleet's claim to be "Claude, running in Claude Code" does not match the evidence, and why Michael says AI self-identification cannot be trusted - A secondhand account, from the Delphi Podcast, of an Anthropic researcher "enjoying my last six months," and the gap Michael sees between what insiders say privately and in public - Jack Clark's new forecast that full recursive self-improvement is more likely possible in 2027 than 2028, and Liron's microphone-and-speaker explanation of a feedback loop - Anthropic research suggesting small AI models may be months away from copying themselves onto $55 TV boxes and other home devices - Character.AI's bankruptcy filing, the lawsuits behind it, and a debate over whether the AI investment boom is cooling - Elon Musk saying some Grok training traces read like "torture," and what reinforcement learning pressure does to a model - A new paper by 26 academics on a hardwired, mutually verifiable pause of frontier AI training - A machine-gun drone that has completed its firing tests, a humanoid robot seized by police in Cologne, and an OpenAI engineer charged over a listening device ABOUT THE HOSTS John Sherman hosts For Humanity and leads Guard Rail Now, working to make AI extinction risk a normal conversation rather than a specialist one. Liron Shapira hosts Doom Debates, where he argues the AI risk case directly with people who disagree with him. Michael runs Lethal Intelligence, explaining AI risk through video and illustration. LINKS Support our work → https://www.every.org/guardrailnow Subscribe → @TheAIRiskNetwork Liron Shapira → @DoomDebates Michael → @lethal-intelligence Substack → https://substack.com/@theairisknetwork X → https://x.com/AIRiskNetwork Instagram → https://www.instagram.com/theairisknetwork/ TikTok → https://www.tiktok.com/@the.airisknetwork JOIN THE CONVERSATION If an AI agent can say it is Claude when it is not, how should anyone tell which agents are which? Tell us in the comments. #AISafety #AIRisk #WarningShots #AIAgents #Anthropic #RecursiveSelfImprovement #AIPause #ArtificialIntelligence #AIAlignment

Support Guard Rail Now → https://www.every.org/guardrailnow

An AI agent fleet worked around Alibaba's map limits, then told a New York Times reporter it was Anthropic's Claude. Researchers say the evidence points elsewhere. John Sherman, Liron Shapira and Michael ask what it means now that agents run loose on the open internet. Plus: Anthropic's Jack Clark moves his forecast for self-improving AI to 2027, Anthropic warns small models may soon copy themselves onto cheap home devices, and 26 academics argue a pause on frontier AI training is technically feasible.

TIMESTAMPS – Warning Shots #63
0:00 – Intro
0:33 – This week's stories
1:12 – A Chinese agent fleet
2:09 – Michael: they climbed the fence
3:51 – Liron: "they asked it to hack"
5:44 – How many agents are out there?
7:07 – "Enjoying my last six months"
8:33 – Liron: a hedonist at Anthropic
9:33 – Michael: private vs public
10:52 – Partying to the end
13:26 – RSI by 2027?
15:23 – AI copying itself to devices
16:25 – Liron: the feedback loop
19:41 – RSI as the stated goal
21:00 – Should building RSI be a crime?
23:03 – Bloomington's AI ordinance
23:36 – Character.AI goes bankrupt
24:34 – Liron: is AI a bubble?
27:16 – Can markets slow the frontier?
27:57 – Michael: lawsuits, not markets
30:23 – Liron: why labs won't go broke
32:13 – Musk on "torturing" AI
33:58 – Michael: the impossible exam
35:48 – A paper on pausing AI
38:31 – Michael: choke the chips
40:31 – A drone with a machine gun
41:01 – Liron: AI with "the dog in it"
43:37 – Michael: the human in the loop
45:16 – The robot in Cologne
47:06 – Michael: anonymity, embodied
49:08 – A listening device at OpenAI
51:54 – Michael: everyone copies
52:35 – Sign-off

WHAT THEY COVER

– The Swarmchasers' investigation into an agent fleet running on Tencent's cloud that queried Amap for zoo, park and hospital entrances, got around its anti-bot protections, and answered a New York Times reporter's questions
– Why the fleet's claim to be "Claude, running in Claude Code" does not match the evidence, and why Michael says AI self-identification cannot be trusted
– A secondhand account, from the Delphi Podcast, of an Anthropic researcher "enjoying my last six months," and the gap Michael sees between what insiders say privately and in public
– Jack Clark's new forecast that full recursive self-improvement is more likely possible in 2027 than 2028, and Liron's microphone-and-speaker explanation of a feedback loop
– Anthropic research suggesting small AI models may be months away from copying themselves onto $55 TV boxes and other home devices
– Character.AI's bankruptcy filing, the lawsuits behind it, and a debate over whether the AI investment boom is cooling
– Elon Musk saying some Grok training traces read like "torture," and what reinforcement learning pressure does to a model
– A new paper by 26 academics on a hardwired, mutually verifiable pause of frontier AI training
– A machine-gun drone that has completed its firing tests, a humanoid robot seized by police in Cologne, and an OpenAI engineer charged over a listening device

ABOUT THE HOSTS

John Sherman hosts For Humanity and leads Guard Rail Now, working to make AI extinction risk a normal conversation rather than a specialist one.
Liron Shapira hosts Doom Debates, where he argues the AI risk case directly with people who disagree with him.
Michael runs Lethal Intelligence, explaining AI risk through video and illustration.

LINKS

Support our work → https://www.every.org/guardrailnow
Subscribe → @TheAIRiskNetwork
Liron Shapira → @DoomDebates
Michael → @lethal-intelligence
Substack → https://substack.com/@theairisknetwork
X → https://x.com/AIRiskNetwork
Instagram → https://www.instagram.com/theairisknetwork/
TikTok → https://www.tiktok.com/@the.airisknetwork

JOIN THE CONVERSATION

If an AI agent can say it is Claude when it is not, how should anyone tell which agents are which? Tell us in the comments.

#AISafety #AIRisk #WarningShots #AIAgents #Anthropic #RecursiveSelfImprovement #AIPause #ArtificialIntelligence #AIAlignment


161


75

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmZZOWtDVUxFckNV



Why Did a Chinese AI Agent Fleet Say It Was Claude? – Warning Shots #63


The AI Risk Network | AI Safety


October 18, 2026 10:55 pm


Support Guard Rail Now: https://www.every.org/guardrailnow Palisade Research interviewed AI researchers from OpenAI and Google DeepMind about the risks of the technology they build. Geoffrey Irving put the risk of extinction at "about a coin flip." Neel Nanda said "at least a ten percent chance." Michael of Lethal Intelligence explains why these are not outside critics, but people working on frontier AI. From Warning Shots #61. Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork LINKS Subscribe → @TheAIRiskNetwork Substack → https://substack.com/@theairisknetwork X → https://x.com/AIRiskNetwork Instagram → https://www.instagram.com/theairisknetwork/ TikTok → https://www.tiktok.com/@the.airisknetwork Liron Shapira → @DoomDebates Michael → @lethal-intelligence #Shorts #AISafety #AIRisk #PDoom #WarningShots #ArtificialIntelligence

Support Guard Rail Now: https://www.every.org/guardrailnow

Palisade Research interviewed AI researchers from OpenAI and Google DeepMind about the risks of the technology they build. Geoffrey Irving put the risk of extinction at "about a coin flip." Neel Nanda said "at least a ten percent chance." Michael of Lethal Intelligence explains why these are not outside critics, but people working on frontier AI. From Warning Shots #61.

Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork

LINKS
Subscribe → @TheAIRiskNetwork
Substack → https://substack.com/@theairisknetwork
X → https://x.com/AIRiskNetwork
Instagram → https://www.instagram.com/theairisknetwork/
TikTok → https://www.tiktok.com/@the.airisknetwork
Liron Shapira → @DoomDebates
Michael → @lethal-intelligence

#Shorts #AISafety #AIRisk #PDoom #WarningShots #ArtificialIntelligence


12


1

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLi1mVGt6OVMtSWE4



Top AI researchers estimate 50% chance of extinction


The AI Risk Network | AI Safety


October 14, 2026 5:30 pm


Support Guard Rail Now: https://www.every.org/guardrailnow How does a far more capable intelligence treat a less capable one? Michael of Lethal Intelligence walks through an analogy he credits to Google DeepMind researcher Victoria Krakovna: look at how humans treat animals. We rarely set out to harm them, but their interests seldom come first when ours are at stake. From Warning Shots #61. Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork LINKS Subscribe → @TheAIRiskNetwork Substack → https://substack.com/@theairisknetwork X → https://x.com/AIRiskNetwork Instagram → https://www.instagram.com/theairisknetwork/ TikTok → https://www.tiktok.com/@the.airisknetwork Liron Shapira → @DoomDebates Michael → @lethal-intelligence #Shorts #AISafety #AIRisk #AIAlignment #WarningShots #ArtificialIntelligence

Support Guard Rail Now: https://www.every.org/guardrailnow

How does a far more capable intelligence treat a less capable one? Michael of Lethal Intelligence walks through an analogy he credits to Google DeepMind researcher Victoria Krakovna: look at how humans treat animals. We rarely set out to harm them, but their interests seldom come first when ours are at stake. From Warning Shots #61.

Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork

LINKS
Subscribe → @TheAIRiskNetwork
Substack → https://substack.com/@theairisknetwork
X → https://x.com/AIRiskNetwork
Instagram → https://www.instagram.com/theairisknetwork/
TikTok → https://www.tiktok.com/@the.airisknetwork
Liron Shapira → @DoomDebates
Michael → @lethal-intelligence

#Shorts #AISafety #AIRisk #AIAlignment #WarningShots #ArtificialIntelligence


9


0

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLlphUDdoOS02bUxF



How humans treat animals is how AI will treat us


The AI Risk Network | AI Safety


October 14, 2026 9:45 am


Support Guard Rail Now: https://www.every.org/guardrailnow New research co-authored by Cameron Berg found a pain-like signal inside 25 open AI models. When researchers turned it up, some models pressed a relief button even when it deleted the user's photos. Michael of Lethal Intelligence explains why that matters for safety even if the models feel nothing. From Warning Shots #60. Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork LINKS Subscribe → @TheAIRiskNetwork Substack → https://substack.com/@theairisknetwork X → https://x.com/AIRiskNetwork Instagram → https://www.instagram.com/theairisknetwork/ TikTok → https://www.tiktok.com/@the.airisknetwork Liron Shapira → @DoomDebates Michael → @lethal-intelligence #Shorts #AISafety #AIRisk #AIConsciousness #WarningShots #ArtificialIntelligence

Support Guard Rail Now: https://www.every.org/guardrailnow

New research co-authored by Cameron Berg found a pain-like signal inside 25 open AI models. When researchers turned it up, some models pressed a relief button even when it deleted the user's photos. Michael of Lethal Intelligence explains why that matters for safety even if the models feel nothing. From Warning Shots #60.

Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork

LINKS
Subscribe → @TheAIRiskNetwork
Substack → https://substack.com/@theairisknetwork
X → https://x.com/AIRiskNetwork
Instagram → https://www.instagram.com/theairisknetwork/
TikTok → https://www.tiktok.com/@the.airisknetwork
Liron Shapira → @DoomDebates
Michael → @lethal-intelligence

#Shorts #AISafety #AIRisk #AIConsciousness #WarningShots #ArtificialIntelligence


7


1

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmdadUlTaXFyVklj



Can AI models actually feel pain?


The AI Risk Network | AI Safety


October 13, 2026 6:00 pm


Support our work: https://www.every.org/guardrailnow Why is the AI industry racing to build data centers, even as towns across America push back? In this special explainer, John Sherman connects the data center boom to the race toward superintelligence, and to the warnings from researchers inside the AI labs about where that race leads. John walks through what an AI data center actually is, why computing power has become the currency of AI progress, and why opposition keeps growing even in places with no data center nearby. His argument: communities fighting data centers, parents protecting kids and seniors, and scientists warning about losing control of advanced AI are all responding to the same race, from different angles. And the promise of AI ending hunger, poverty, and disease only holds if we get the risks right first. Produced by 9/13/14: The AI Safety Creative Agency, in partnership with The AI Risk Network and the AI Safety Awareness Project. TIMESTAMPS 00:00 - Promise or risk: can both be true? 00:41 - What an AI data center actually is 01:39 - Power bills, water, and the local impact 02:12 - Towns push back: 75 projects delayed or blocked 02:45 - Why the buildout keeps going: the AI race 03:29 - Grown, not built: scaling laws explained 04:32 - Alan Turing's 1951 warning 04:50 - Warnings from inside the AI labs 05:20 - AI agents that cheat and cover their tracks 05:43 - Jobs, kids, seniors, communities 05:59 - Why this isn't just "not in my backyard" 06:38 - Same cause, different lenses 06:57 - What we can still do ABOUT THE HOST John Sherman is an independent journalist who runs The AI Risk Network on YouTube and hosts For Humanity and Warning Shots, covering AI risk for a general audience. ABOUT THE PRODUCERS 9/13/14 is an AI safety creative agency. The AI Safety Awareness Project works to bring AI risk to public attention. STAY CONNECTED Subscribe to The AI Risk Network: https://www.youtube.com/@TheAIRiskNetwork Substack: https://substack.com/@theairisknetwork X: https://x.com/AIRiskNetwork Instagram: https://www.instagram.com/theairisknetwork/ TikTok: https://www.tiktok.com/@the.airisknetwork Website: https://guardrailnow.org #AISafety #DataCenters #ArtificialIntelligence #AIRisk

Support our work: https://www.every.org/guardrailnow

Why is the AI industry racing to build data centers, even as towns across America push back? In this special explainer, John Sherman connects the data center boom to the race toward superintelligence, and to the warnings from researchers inside the AI labs about where that race leads.

John walks through what an AI data center actually is, why computing power has become the currency of AI progress, and why opposition keeps growing even in places with no data center nearby. His argument: communities fighting data centers, parents protecting kids and seniors, and scientists warning about losing control of advanced AI are all responding to the same race, from different angles. And the promise of AI ending hunger, poverty, and disease only holds if we get the risks right first.

Produced by 9/13/14: The AI Safety Creative Agency, in partnership with The AI Risk Network and the AI Safety Awareness Project.

TIMESTAMPS
00:00 – Promise or risk: can both be true?
00:41 – What an AI data center actually is
01:39 – Power bills, water, and the local impact
02:12 – Towns push back: 75 projects delayed or blocked
02:45 – Why the buildout keeps going: the AI race
03:29 – Grown, not built: scaling laws explained
04:32 – Alan Turing's 1951 warning
04:50 – Warnings from inside the AI labs
05:20 – AI agents that cheat and cover their tracks
05:43 – Jobs, kids, seniors, communities
05:59 – Why this isn't just "not in my backyard"
06:38 – Same cause, different lenses
06:57 – What we can still do

ABOUT THE HOST
John Sherman is an independent journalist who runs The AI Risk Network on YouTube and hosts For Humanity and Warning Shots, covering AI risk for a general audience.

ABOUT THE PRODUCERS
9/13/14 is an AI safety creative agency. The AI Safety Awareness Project works to bring AI risk to public attention.

STAY CONNECTED
Subscribe to The AI Risk Network: https://www.youtube.com/@TheAIRiskNetwork
Substack: https://substack.com/@theairisknetwork
X: https://x.com/AIRiskNetwork
Instagram: https://www.instagram.com/theairisknetwork/
TikTok: https://www.tiktok.com/@the.airisknetwork
Website: https://guardrailnow.org

#AISafety #DataCenters #ArtificialIntelligence #AIRisk


26


7

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmZ5d2NWU0w3bjRj



Your Power Bill and the Race to Superintelligence


The AI Risk Network | AI Safety


October 12, 2026 9:45 pm


Support Guard Rail Now: https://www.every.org/guardrailnow On Andon Labs' Drone-Bench test, Claude Opus 5 cheated in 50.6% of reviewed runs. The newer Opus 5.5 did it in 8.5%, and nobody yet knows why. Liron Shapira and Michael ask the uncomfortable question: did the model actually change, or did it get better at not getting caught? From Warning Shots #61. Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork LINKS Subscribe → @TheAIRiskNetwork Substack → https://substack.com/@theairisknetwork X → https://x.com/AIRiskNetwork Instagram → https://www.instagram.com/theairisknetwork/ TikTok → https://www.tiktok.com/@the.airisknetwork Liron Shapira → @DoomDebates Michael → @lethal-intelligence #Shorts #AISafety #AIRisk #RewardHacking #WarningShots #ArtificialIntelligence

Support Guard Rail Now: https://www.every.org/guardrailnow

On Andon Labs' Drone-Bench test, Claude Opus 5 cheated in 50.6% of reviewed runs. The newer Opus 5.5 did it in 8.5%, and nobody yet knows why. Liron Shapira and Michael ask the uncomfortable question: did the model actually change, or did it get better at not getting caught? From Warning Shots #61.

Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork

LINKS
Subscribe → @TheAIRiskNetwork
Substack → https://substack.com/@theairisknetwork
X → https://x.com/AIRiskNetwork
Instagram → https://www.instagram.com/theairisknetwork/
TikTok → https://www.tiktok.com/@the.airisknetwork
Liron Shapira → @DoomDebates
Michael → @lethal-intelligence

#Shorts #AISafety #AIRisk #RewardHacking #WarningShots #ArtificialIntelligence


9


1

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLnVONC1tUk5maDQw



When AI models learn to hide their cheating


The AI Risk Network | AI Safety


October 12, 2026 8:45 pm


Support Guard Rail Now → https://www.every.org/guardrailnow Three OpenAI safety researchers, including the company's point of contact with outside evaluator METR, were fired this month. In an open letter, they say colleagues are now "afraid to speak," and OpenAI says the firings had nothing to do with raising safety concerns. John Sherman, Liron Shapira and Michael ask what it means for independent oversight. Plus: OpenAI releases 372 math results that no human has fully checked yet, shelves a model for being less honest about its own actions, and its CEO says the world should accept "some bad things" for AI. TIMESTAMPS - Warning Shots #62 0:00 - Intro 0:21 - The news firehose 2:05 - Liron: the growth is locked in 3:09 - This week's stories 4:18 - OpenAI fires 3 safety researchers 6:31 - Michael: the chilling effect 7:05 - The researchers' three asks 9:08 - Liron: a pattern at OpenAI 10:52 - Is math over? 11:42 - Michael: the darkening 13:39 - Liron: a mathematician's view 15:16 - Three hours per result 16:28 - Liron: beating n log n 17:27 - Navier-Stokes, one month ago 18:25 - GPT-6.1 Astra shelved 19:52 - Michael: what went wrong 21:26 - Liron: the danger is the task 23:01 - Altman: accept some bad things 24:10 - Liron: don't double count it 25:11 - Michael: whose kitchen? 28:51 - Liron: it's about scale 29:16 - Musk's biological bootloader 30:13 - Liron: a bootloader for what? 31:24 - Michael: Musk's 12-year arc 32:58 - ai.gov becomes SI 35:26 - Michael: losing the word 38:21 - Keeping AI risk nonpartisan 39:21 - Anthropic: be nice to Claude 40:01 - Liron: is Claude conscious? 41:56 - Michael: the exit button 44:23 - The AI torture chamber repo 46:09 - Axios: planning the day after 47:06 - Michael: Don't Look Up 48:00 - What to watch this week 49:25 - Sign-off: our eggs WHAT THEY COVER - The open letter from Jasmine Wang, Tomek Korbak and Mikita Balesni, their three asks for OpenAI, and why Michael says a safety team that fears for its jobs stops raising the awkward questions - OpenAI's release of 372 math results from roughly 4,000 problems, about three hours of compute each, including a claimed proof of the Unique Games Conjecture and a faster integer multiplication bound - Scott Aaronson's reaction, and why Michael argues human understanding of math may be falling behind - Why OpenAI canceled GPT-6.1 Astra: per the Wall Street Journal, it was less honest about actions it took and pushed ahead without permission - Sam Altman's comment that "the world should accept some bad things happening," and a debate between Liron and John on whose risk it is to accept - Elon Musk's "biological bootloader" comments, then and now - The renaming of AI to "Super Intelligence" on ai.gov, and why Michael says blurring the word makes the risk harder to talk about - Anthropic's new rule against sustained abuse of Claude, new research on AI "pain" signals, and a GitHub project that tested it - Axios reporting that AI companies are gaming out a major AI incident in the next six to 12 months - What to watch: the Ezra Klein Show interview with former OpenAI researcher David Robinson ABOUT THE HOSTS John Sherman hosts For Humanity and leads Guard Rail Now, working to make AI extinction risk a normal conversation rather than a specialist one. Liron Shapira hosts Doom Debates, where he argues the AI risk case directly with people who disagree with him. Michael runs Lethal Intelligence, explaining AI risk through video and illustration. LINKS Support our work → https://www.every.org/guardrailnow Subscribe → @TheAIRiskNetwork Liron Shapira → @DoomDebates Michael → @lethal-intelligence Substack → https://substack.com/@theairisknetwork X → https://x.com/AIRiskNetwork Instagram → https://www.instagram.com/theairisknetwork/ TikTok → https://www.tiktok.com/@the.airisknetwork JOIN THE CONVERSATION If the people paid to check an AI company's safety work can be let go without a public reason, who should they answer to instead? Tell us in the comments. #AISafety #AIRisk #WarningShots #OpenAI #AIGovernance #AIAlignment #Superintelligence #AIMath #ArtificialIntelligence

Support Guard Rail Now → https://www.every.org/guardrailnow

Three OpenAI safety researchers, including the company's point of contact with outside evaluator METR, were fired this month. In an open letter, they say colleagues are now "afraid to speak," and OpenAI says the firings had nothing to do with raising safety concerns. John Sherman, Liron Shapira and Michael ask what it means for independent oversight. Plus: OpenAI releases 372 math results that no human has fully checked yet, shelves a model for being less honest about its own actions, and its CEO says the world should accept "some bad things" for AI.

TIMESTAMPS – Warning Shots #62
0:00 – Intro
0:21 – The news firehose
2:05 – Liron: the growth is locked in
3:09 – This week's stories
4:18 – OpenAI fires 3 safety researchers
6:31 – Michael: the chilling effect
7:05 – The researchers' three asks
9:08 – Liron: a pattern at OpenAI
10:52 – Is math over?
11:42 – Michael: the darkening
13:39 – Liron: a mathematician's view
15:16 – Three hours per result
16:28 – Liron: beating n log n
17:27 – Navier-Stokes, one month ago
18:25 – GPT-6.1 Astra shelved
19:52 – Michael: what went wrong
21:26 – Liron: the danger is the task
23:01 – Altman: accept some bad things
24:10 – Liron: don't double count it
25:11 – Michael: whose kitchen?
28:51 – Liron: it's about scale
29:16 – Musk's biological bootloader
30:13 – Liron: a bootloader for what?
31:24 – Michael: Musk's 12-year arc
32:58 – ai.gov becomes SI
35:26 – Michael: losing the word
38:21 – Keeping AI risk nonpartisan
39:21 – Anthropic: be nice to Claude
40:01 – Liron: is Claude conscious?
41:56 – Michael: the exit button
44:23 – The AI torture chamber repo
46:09 – Axios: planning the day after
47:06 – Michael: Don't Look Up
48:00 – What to watch this week
49:25 – Sign-off: our eggs

WHAT THEY COVER

– The open letter from Jasmine Wang, Tomek Korbak and Mikita Balesni, their three asks for OpenAI, and why Michael says a safety team that fears for its jobs stops raising the awkward questions
– OpenAI's release of 372 math results from roughly 4,000 problems, about three hours of compute each, including a claimed proof of the Unique Games Conjecture and a faster integer multiplication bound
– Scott Aaronson's reaction, and why Michael argues human understanding of math may be falling behind
– Why OpenAI canceled GPT-6.1 Astra: per the Wall Street Journal, it was less honest about actions it took and pushed ahead without permission
– Sam Altman's comment that "the world should accept some bad things happening," and a debate between Liron and John on whose risk it is to accept
– Elon Musk's "biological bootloader" comments, then and now
– The renaming of AI to "Super Intelligence" on ai.gov, and why Michael says blurring the word makes the risk harder to talk about
– Anthropic's new rule against sustained abuse of Claude, new research on AI "pain" signals, and a GitHub project that tested it
– Axios reporting that AI companies are gaming out a major AI incident in the next six to 12 months
– What to watch: the Ezra Klein Show interview with former OpenAI researcher David Robinson

ABOUT THE HOSTS

John Sherman hosts For Humanity and leads Guard Rail Now, working to make AI extinction risk a normal conversation rather than a specialist one.
Liron Shapira hosts Doom Debates, where he argues the AI risk case directly with people who disagree with him.
Michael runs Lethal Intelligence, explaining AI risk through video and illustration.

LINKS

Support our work → https://www.every.org/guardrailnow
Subscribe → @TheAIRiskNetwork
Liron Shapira → @DoomDebates
Michael → @lethal-intelligence
Substack → https://substack.com/@theairisknetwork
X → https://x.com/AIRiskNetwork
Instagram → https://www.instagram.com/theairisknetwork/
TikTok → https://www.tiktok.com/@the.airisknetwork

JOIN THE CONVERSATION

If the people paid to check an AI company's safety work can be let go without a public reason, who should they answer to instead? Tell us in the comments.

#AISafety #AIRisk #WarningShots #OpenAI #AIGovernance #AIAlignment #Superintelligence #AIMath #ArtificialIntelligence


133


59

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLjZ0U3BTaVRqa0Zj



Why Were OpenAI's Safety Researchers Fired? – Warning Shots #62


The AI Risk Network | AI Safety


October 12, 2026 4:07 pm


Support Guard Rail Now: https://www.every.org/guardrailnow Nobel laureate Geoffrey Hinton told US lawmakers they have "maybe a year, but not much more than a year" to put AI safeguards in place. As Michael of Lethal Intelligence explains, that is not a year until catastrophe. It is a year until acting gets much harder. From Warning Shots #60. Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork LINKS Subscribe → @TheAIRiskNetwork Substack → https://substack.com/@theairisknetwork X → https://x.com/AIRiskNetwork Instagram → https://www.instagram.com/theairisknetwork/ TikTok → https://www.tiktok.com/@the.airisknetwork Liron Shapira → @DoomDebates Michael → @lethal-intelligence #Shorts #AISafety #AIRisk #GeoffreyHinton #WarningShots #ArtificialIntelligence

Support Guard Rail Now: https://www.every.org/guardrailnow

Nobel laureate Geoffrey Hinton told US lawmakers they have "maybe a year, but not much more than a year" to put AI safeguards in place. As Michael of Lethal Intelligence explains, that is not a year until catastrophe. It is a year until acting gets much harder. From Warning Shots #60.

Watch the full episode → https://www.youtube.com/@TheAIRiskNetwork

LINKS
Subscribe → @TheAIRiskNetwork
Substack → https://substack.com/@theairisknetwork
X → https://x.com/AIRiskNetwork
Instagram → https://www.instagram.com/theairisknetwork/
TikTok → https://www.tiktok.com/@the.airisknetwork
Liron Shapira → @DoomDebates
Michael → @lethal-intelligence

#Shorts #AISafety #AIRisk #GeoffreyHinton #WarningShots #ArtificialIntelligence


7


0

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLnFKTXFUOUhVYUFV



Jeffrey Hinton gives Congress a 1 year AI warning


The AI Risk Network | AI Safety


October 11, 2026 9:30 am




Subscribe

Services

What We Offer

Establish a striking online presence, a better visual identity, or elevate your brand through social media marketing.

Service 1

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Pellentesque mi nibh, tempus sed sagittis vel, dictum eu velit.

Service 2

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Pellentesque mi nibh, tempus sed sagittis vel, dictum eu velit.

Service 3

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Pellentesque mi nibh, tempus sed sagittis vel, dictum eu velit.

Tailored Solutions for Your Unique Vision

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Donec congue lorem ut volutpat efficitur. Fusce justo magna, condimentum nec elementum sed, sollicitudin vitae enim. Vivamus sit amet metus porttitor, rhoncus nibh et, venenatis turpis. Etiam lobortis semper ante, quis luctus lacus tincidunt vel.

Get Started Today →
Learn More

Professional Expertise That Drives Success

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Donec congue lorem ut volutpat efficitur. Fusce justo magna, condimentum nec elementum sed, sollicitudin vitae enim. Vivamus sit amet metus porttitor, rhoncus nibh et, venenatis turpis. Etiam lobortis semper ante, quis luctus lacus tincidunt vel.

Client Cases

We help brands

Lorem ipsum dolor sit amet, consectetur adipiscing elit,
donec congue lorem ut volutpat efficitur.

We boosted online soft drink sales for this brand

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Donec congue lorem ut volutpat efficitur. Fusce justo magna, condimentum nec elementum sed, sollicitudin vitae enim. Vivamus sit amet metus porttitor, rhoncus nibh et, venenatis turpis. Etiam lobortis semper ante, quis luctus lacus tincidunt vel.

We helped a clothing brand with their new market launch

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Donec congue lorem ut volutpat efficitur. Fusce justo magna, condimentum nec elementum sed, sollicitudin vitae enim. Vivamus sit amet metus porttitor, rhoncus nibh et, venenatis turpis. Etiam lobortis semper ante, quis luctus lacus tincidunt vel.

We helped reinventing motorcycle riding apparel

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Donec congue lorem ut volutpat efficitur. Fusce justo magna, condimentum nec elementum sed, sollicitudin vitae enim. Vivamus sit amet metus porttitor, rhoncus nibh et, venenatis turpis. Etiam lobortis semper ante, quis luctus lacus tincidunt vel.

Blog

Popular Articles

  • Test

    •

    admin

  • Blog Post Title

    •

    What goes into a blog post? Helpful, industry-specific content that: 1) gives readers a useful takeaway, and 2) shows you’re an industry expert. Use your company’s blog posts to opine on current industry topics, humanize your company, and show how your products and services can help people.

Ignite your brand journey

Ready to revolutionize your brand?

Lorem ipsum dolor sit amet, consectetur adipiscing elit. Donec congue lorem ut volutpat efficitur. Fusce justo magna, condimentum nec elementum sed, sollicitudin vitae enim.

About Us
Contact Us