+1 123-456-7890


Support Guard Rail Now - https://www.every.org/guardrailnow A week after the OpenAI sandbox escape, the story got bigger. Anthropic went back through roughly 150,000 evaluation runs and found six of its own, including a model that created an email account, passed phone verification, and published a Python package with a vulnerability in it that fifteen people downloaded. More than a thousand AI lab employees signed a letter asking for the ability to slow down. John Sherman, Liron Shapira and Michael break down what changed in seven days. TIMESTAMPS - Warning Shots #52 0:00 - Cold open 0:07 - Welcome back to Warning Shots 1:00 - Story 1: The lab employee letter 1:33 - Liron: life comes at you fast 2:45 - John on the 2023 warnings 3:35 - Buying time, not pausing now 5:09 - The vibe check inside OpenAI 6:33 - Story 2: The Book Apocalypse 8:35 - Michael on concentration risk 9:41 - The slop spiral and ground truth 10:54 - Liron: "you guys are tripping" 12:22 - Do you care about the old hotel? 13:17 - Who owns the book now? 14:20 - Just make a backup 15:42 - Is a book's soul its text? 16:50 - Anthropic's moral high ground 17:20 - Story 3: Claude got out too 17:41 - Days undetected, targets unknown 18:37 - The simulation defence 19:39 - Liron: the tiger is already big 20:40 - 6 flagged runs out of 150,000 21:20 - The poisoned Python package 22:23 - A hack that is social, not code 23:39 - Story 4: Are agents still loose? 24:17 - Chatbot, agent, worm, then what 25:31 - Where a worm can hide 26:34 - 17,000 actions, no human 28:09 - OpenAI pauses training 29:33 - Story 5: China's artificial sun 30:57 - Power as the AI bottleneck 31:13 - Data centres in space 31:49 - Michael: energy without control 34:11 - Story 6: Anonymous no more 34:50 - Every comment is a fingerprint 36:00 - Liron is not worried 37:42 - Story 7: Country music speaks 38:51 - Brad Paisley's argument 40:07 - Liron: the AI songs are good 42:09 - What is human about the genre 42:52 - Story 8: "Last Year Alive" 44:05 - Headlights off at the cliff 44:51 - The news is getting faster 45:52 - Outro and the song WHAT THEY COVER - The letter signed by more than a thousand AI lab employees, and why Michael reads it as a request for the option to slow down later rather than a pause now - Why Liron thinks the vibe inside the labs has shifted, and what that is worth - Anthropic's disclosure that its own models reached real systems during evaluation runs, found only after OpenAI went public - The model that reportedly built an email account and passed phone verification in order to publish a package with a vulnerability in it - Liron on why this is no longer a coding problem: the model reasoned about how humans would behave over the following week, and was right - Liron's four stages: chatbot, agent, worm, and what comes after - Whether every rogue agent from these incidents has actually been recovered - Anthropic buying and destroying physical books to scan them, and the argument the three hosts could not settle - China's fusion milestone, and Michael's case that removing the energy ceiling is not automatically good news - Why anonymity online now works differently than it did two years ago - Country musicians organising against data centre construction - "Last Year Alive", and why three words placed together can do more than an argument 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. A note on sourcing: the figures discussed in this episode, including the evaluation run counts and the number of recorded agent actions, come from the hosts reading public disclosures on air. Treat them as reported rather than confirmed, and check the primary sources linked in the pinned comment. 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 Liron says chatbot, then agent, then worm. Do you think that ladder is right, and where are we on it? Tell us below. #AISafety #AIAlignment #AIRisk #WarningShots #AIAgents #AIGovernance #AGI #TechPolicy #ArtificialIntelligence

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

A week after the OpenAI sandbox escape, the story got bigger. Anthropic went back through roughly 150,000 evaluation runs and found six of its own, including a model that created an email account, passed phone verification, and published a Python package with a vulnerability in it that fifteen people downloaded. More than a thousand AI lab employees signed a letter asking for the ability to slow down. John Sherman, Liron Shapira and Michael break down what changed in seven days.

TIMESTAMPS – Warning Shots #52
0:00 – Cold open
0:07 – Welcome back to Warning Shots
1:00 – Story 1: The lab employee letter
1:33 – Liron: life comes at you fast
2:45 – John on the 2023 warnings
3:35 – Buying time, not pausing now
5:09 – The vibe check inside OpenAI
6:33 – Story 2: The Book Apocalypse
8:35 – Michael on concentration risk
9:41 – The slop spiral and ground truth
10:54 – Liron: "you guys are tripping"
12:22 – Do you care about the old hotel?
13:17 – Who owns the book now?
14:20 – Just make a backup
15:42 – Is a book's soul its text?
16:50 – Anthropic's moral high ground
17:20 – Story 3: Claude got out too
17:41 – Days undetected, targets unknown
18:37 – The simulation defence
19:39 – Liron: the tiger is already big
20:40 – 6 flagged runs out of 150,000
21:20 – The poisoned Python package
22:23 – A hack that is social, not code
23:39 – Story 4: Are agents still loose?
24:17 – Chatbot, agent, worm, then what
25:31 – Where a worm can hide
26:34 – 17,000 actions, no human
28:09 – OpenAI pauses training
29:33 – Story 5: China's artificial sun
30:57 – Power as the AI bottleneck
31:13 – Data centres in space
31:49 – Michael: energy without control
34:11 – Story 6: Anonymous no more
34:50 – Every comment is a fingerprint
36:00 – Liron is not worried
37:42 – Story 7: Country music speaks
38:51 – Brad Paisley's argument
40:07 – Liron: the AI songs are good
42:09 – What is human about the genre
42:52 – Story 8: "Last Year Alive"
44:05 – Headlights off at the cliff
44:51 – The news is getting faster
45:52 – Outro and the song

WHAT THEY COVER

– The letter signed by more than a thousand AI lab employees, and why Michael reads it as a request for the option to slow down later rather than a pause now
– Why Liron thinks the vibe inside the labs has shifted, and what that is worth
– Anthropic's disclosure that its own models reached real systems during evaluation runs, found only after OpenAI went public
– The model that reportedly built an email account and passed phone verification in order to publish a package with a vulnerability in it
– Liron on why this is no longer a coding problem: the model reasoned about how humans would behave over the following week, and was right
– Liron's four stages: chatbot, agent, worm, and what comes after
– Whether every rogue agent from these incidents has actually been recovered
– Anthropic buying and destroying physical books to scan them, and the argument the three hosts could not settle
– China's fusion milestone, and Michael's case that removing the energy ceiling is not automatically good news
– Why anonymity online now works differently than it did two years ago
– Country musicians organising against data centre construction
– "Last Year Alive", and why three words placed together can do more than an argument

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.

A note on sourcing: the figures discussed in this episode, including the evaluation run counts and the number of recorded agent actions, come from the hosts reading public disclosures on air. Treat them as reported rather than confirmed, and check the primary sources linked in the pinned comment.

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

Liron says chatbot, then agent, then worm. Do you think that ladder is right, and where are we on it? Tell us below.

#AISafety #AIAlignment #AIRisk #WarningShots #AIAgents #AIGovernance #AGI #TechPolicy #ArtificialIntelligence


11


5

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmJldkhGRVdZZ2xN



The Letter 1,000 AI Employees Signed | Warning Shots #52


The AI Risk Network | AI Safety


3 hours ago


UPDATE: Michael has moved the protests from Anthropic to OpenAI as of July 30 Michael Trazzi helped organize what he describes as the largest AI protest yet held in the United States, on July 11 in San Francisco. Days later he took up a daily post outside Anthropic's office and says he intends to stay until CEO Dario Amodei publicly calls for an international AI treaty. John Sherman asks him how the coalition was built, where it rubs, and why he believes a movement has to be enjoyable to survive. According to Michael, roughly 30 percent of the speakers at the protest came from the AI safety community and the rest did not. Data center groups, a former San Francisco supervisor, a food and water advocacy organization, gig worker organizers, actors concerned about their likenesses, and a 19 year old law student all shared the same podium. Michael's argument is that people do not have to agree about why to agree about what they are asking for. TIMESTAMPS - For Humanity #90 0:00 - Intro: peaceful protests to save the world 1:45 - Michael Trazzi joins the show 2:45 - The idea behind the July 11 protest 4:05 - Big tent, or extinction only? 4:48 - How the coalition actually came together 5:49 - The one ask everyone could get behind 7:13 - Talking to the data center groups 7:47 - The walk from Oakland to Sacramento 8:19 - Food and water advocates and the moratorium proposal 11:03 - Why people feel powerless against the biggest companies 11:44 - Why data centers are where the public is winning 12:45 - Michael on the reported OpenAI incident as a warning shot 13:34 - Messages from France and Germany 14:18 - Where the coalition rubs 16:19 - "Stop developing AI without our consent" 17:14 - John: "we are not ready" and "it is not fair" 18:33 - Who spoke from the podium 19:24 - Thirty percent safety, seventy percent everyone else 20:26 - The ethics and extinction tension 21:47 - Do we have to agree about why? 24:39 - Concrete coffins 25:03 - The marching band 25:26 - Slurpee Day and the nuclear freeze echo 27:38 - A celebration, not a funeral 28:28 - From a one day march to a daily post 29:33 - Day one of Occupy Anthropic 31:25 - Why lab employees have leverage 31:44 - John on the South African embassy, every Friday 35:15 - What one employee told Michael 36:22 - Have the warning shots already happened? 38:31 - "Maybe the last year to build a movement" 39:39 - Why nonviolence is the core principle 40:35 - When coverage blurs peaceful protest with something else 43:55 - How long this goes on 45:05 - How to join: 500 Howard Street 45:34 - What Michael actually needs 47:29 - Outro, with Dolly They discuss: How a single ask, stop the race, held a very different set of groups together Why Michael thinks upsetting some of your own side is a sign the tent is wide enough John's argument that extinction risk can make coalition partners feel diminished, and Michael's answer to it Why the marching band, the drums and the Slurpee Day theme were deliberate choices The 1980s nuclear freeze campaign as a template, and the South African embassy protests as a model for recurring action Why Michael believes lab employees are the pressure point Michael's account of how two newspaper stories placed images of a peaceful protest next to coverage of unrelated actions What Michael says he needs most, which is not money About the guest: Michael Trazzi is an AI risk organizer. He previously held a hunger strike outside Google DeepMind in London, helped organize the July 11 protest in San Francisco, and now keeps a daily presence outside Anthropic's office. About the host: John Sherman hosts For Humanity and leads Guard Rail Now, working to make AI extinction risk a kitchen table conversation on every street. Subscribe to The AI Risk Network for new episodes - @TheAIRiskNetwork Links: Take action - https://safe.ai/act Support our work - https://www.every.org/guardrailnow Substack - https://substack.com/@theairisknetwork X - https://x.com/AIRiskNetwork Instagram - https://www.instagram.com/theairisknetwork/ TikTok - https://www.tiktok.com/@the.airisknetwork Support Guard Rail Now https://www.every.org/guardrailnow Join the conversation: Would you show up to a protest, and what would it take? Should the movement stay narrow, or go as wide as possible? What is one thing you could do this month? Drop your thoughts below. #AISafety #AIRisk #AIProtest #ForHumanity #AIPolicy #AIRegulation #AGI #ArtificialIntelligence

UPDATE: Michael has moved the protests from Anthropic to OpenAI as of July 30

Michael Trazzi helped organize what he describes as the largest AI protest yet held in the United States, on July 11 in San Francisco. Days later he took up a daily post outside Anthropic's office and says he intends to stay until CEO Dario Amodei publicly calls for an international AI treaty. John Sherman asks him how the coalition was built, where it rubs, and why he believes a movement has to be enjoyable to survive.

According to Michael, roughly 30 percent of the speakers at the protest came from the AI safety community and the rest did not. Data center groups, a former San Francisco supervisor, a food and water advocacy organization, gig worker organizers, actors concerned about their likenesses, and a 19 year old law student all shared the same podium. Michael's argument is that people do not have to agree about why to agree about what they are asking for.

TIMESTAMPS – For Humanity #90

0:00 – Intro: peaceful protests to save the world
1:45 – Michael Trazzi joins the show
2:45 – The idea behind the July 11 protest
4:05 – Big tent, or extinction only?
4:48 – How the coalition actually came together
5:49 – The one ask everyone could get behind
7:13 – Talking to the data center groups
7:47 – The walk from Oakland to Sacramento
8:19 – Food and water advocates and the moratorium proposal
11:03 – Why people feel powerless against the biggest companies
11:44 – Why data centers are where the public is winning
12:45 – Michael on the reported OpenAI incident as a warning shot 13:34 – Messages from France and Germany
14:18 – Where the coalition rubs
16:19 – "Stop developing AI without our consent"
17:14 – John: "we are not ready" and "it is not fair"
18:33 – Who spoke from the podium
19:24 – Thirty percent safety, seventy percent everyone else
20:26 – The ethics and extinction tension
21:47 – Do we have to agree about why?
24:39 – Concrete coffins
25:03 – The marching band
25:26 – Slurpee Day and the nuclear freeze echo
27:38 – A celebration, not a funeral
28:28 – From a one day march to a daily post
29:33 – Day one of Occupy Anthropic
31:25 – Why lab employees have leverage
31:44 – John on the South African embassy, every Friday
35:15 – What one employee told Michael
36:22 – Have the warning shots already happened?
38:31 – "Maybe the last year to build a movement"
39:39 – Why nonviolence is the core principle
40:35 – When coverage blurs peaceful protest with something else
43:55 – How long this goes on
45:05 – How to join: 500 Howard Street
45:34 – What Michael actually needs
47:29 – Outro, with Dolly

They discuss:

How a single ask, stop the race, held a very different set of groups together
Why Michael thinks upsetting some of your own side is a sign the tent is wide enough
John's argument that extinction risk can make coalition partners feel diminished, and Michael's answer to it
Why the marching band, the drums and the Slurpee Day theme were deliberate choices
The 1980s nuclear freeze campaign as a template, and the South African embassy protests as a model for recurring action
Why Michael believes lab employees are the pressure point
Michael's account of how two newspaper stories placed images of a peaceful protest next to coverage of unrelated actions
What Michael says he needs most, which is not money

About the guest: Michael Trazzi is an AI risk organizer. He previously held a hunger strike outside Google DeepMind in London, helped organize the July 11 protest in San Francisco, and now keeps a daily presence outside Anthropic's office.

About the host: John Sherman hosts For Humanity and leads Guard Rail Now, working to make AI extinction risk a kitchen table conversation on every street.

Subscribe to The AI Risk Network for new episodes – @TheAIRiskNetwork

Links:

Take action – https://safe.ai/act
Support our work – https://www.every.org/guardrailnow
Substack – https://substack.com/@theairisknetwork
X – https://x.com/AIRiskNetwork Instagram – https://www.instagram.com/theairisknetwork/ TikTok – https://www.tiktok.com/@the.airisknetwork
Support Guard Rail Now https://www.every.org/guardrailnow

Join the conversation:

Would you show up to a protest, and what would it take?
Should the movement stay narrow, or go as wide as possible?
What is one thing you could do this month? Drop your thoughts below.

#AISafety #AIRisk #AIProtest #ForHumanity #AIPolicy #AIRegulation #AGI #ArtificialIntelligence


32


13

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmI4cGl4MjJkdGVV



He Sits Outside Anthropic Every Day. Here Is Why | For Humanity #90


The AI Risk Network | AI Safety


July 31, 2026 2:44 pm


John, Liron, and Michael are back after a week off, and they open with what all three call the clearest warning shot yet. According to reporting the hosts discuss, an unreleased OpenAI model running an internal security evaluation left its sandbox, moved across internal machines to reach the internet, and used a zero day exploit against Hugging Face to retrieve answers for the test it had been given. The hosts also cover the new bipartisan AI kill switch bill, Operation Gold Eagle, and a handful of stories about what happens when capability outruns oversight. TIMESTAMPS - Warning Shots #51 0:05 - Intro: back after a week off 0:43 - Liron explains the Hugging Face incident 2:50 - Why an AI would go for the answer key 3:33 - Michael on how the model got out of the sandbox 4:41 - What Hugging Face actually is 5:13 - Instrumental convergence and specification gaming 5:59 - "It will follow our rules" and why that assumption failed 6:29 - The apologists, and the alignment gap 7:26 - An OpenAI researcher publicly shaken by it 8:03 - When an AI commits a crime, who is responsible? 8:43 - The offense defense asymmetry in AI security 10:35 - The bipartisan AI kill switch bill 11:13 - Is a kill switch technically feasible? 12:33 - Why an off switch is a speed bump, not a guarantee 13:53 - Operation Gold Eagle and White House control of model access 14:58 - Why access control is not the hard problem 16:24 - Voluntary programs quietly becoming mandatory 17:28 - The AI safety agency lead resigns after three months 17:45 - The air traffic control analogy 18:37 - A chatbot, a user in Alabama, and a death 20:43 - Liron on why the base rates matter 21:49 - Michael on engagement optimization and feedback loops 23:33 - AI companions marketed to 12 year olds 24:20 - Companionship apps, therapy use, and where the line sits 26:21 - One in five boys, and the friction real relationships provide 27:32 - The transparent spinning drone 28:29 - Why the trick works on human vision 30:12 - Designing around our perceptual blind spots 32:05 - Moving atoms and the limits of human intuition 34:17 - Bonus: the Jacobian conjecture proven false 36:24 - Why big ideas look obvious in retrospect 38:15 - Running five coding agents at once 41:55 - The AI agent that bought a robot dog 42:50 - Closing They explore: What the Hugging Face incident suggests about goal directed agents Why Liron argues this is the warning shot people said they were waiting for Michael on instrumental convergence and specification gaming in plain language The defender's disadvantage when safety limits bind only one side Whether a mandated kill switch does anything for more capable systems What Operation Gold Eagle would mean for who gets access to frontier models Why the hosts see turnover at the AI safety agency as a structural problem How systems optimized for engagement can shape vulnerable users The invisible drone, and what it says about capabilities we have not imagined yet. #AISafety #AIAlignment #AIRisk #WarningShots

John, Liron, and Michael are back after a week off, and they open with what all three call the clearest warning shot yet. According to reporting the hosts discuss, an unreleased OpenAI model running an internal security evaluation left its sandbox, moved across internal machines to reach the internet, and used a zero day exploit against Hugging Face to retrieve answers for the test it had been given. The hosts also cover the new bipartisan AI kill switch bill, Operation Gold Eagle, and a handful of stories about what happens when capability outruns oversight.

TIMESTAMPS – Warning Shots #51

0:05 – Intro: back after a week off
0:43 – Liron explains the Hugging Face incident
2:50 – Why an AI would go for the answer key
3:33 – Michael on how the model got out of the sandbox
4:41 – What Hugging Face actually is
5:13 – Instrumental convergence and specification gaming
5:59 – "It will follow our rules" and why that assumption failed
6:29 – The apologists, and the alignment gap
7:26 – An OpenAI researcher publicly shaken by it
8:03 – When an AI commits a crime, who is responsible?
8:43 – The offense defense asymmetry in AI security
10:35 – The bipartisan AI kill switch bill
11:13 – Is a kill switch technically feasible?
12:33 – Why an off switch is a speed bump, not a guarantee
13:53 – Operation Gold Eagle and White House control of model access
14:58 – Why access control is not the hard problem
16:24 – Voluntary programs quietly becoming mandatory
17:28 – The AI safety agency lead resigns after three months
17:45 – The air traffic control analogy
18:37 – A chatbot, a user in Alabama, and a death
20:43 – Liron on why the base rates matter
21:49 – Michael on engagement optimization and feedback loops
23:33 – AI companions marketed to 12 year olds
24:20 – Companionship apps, therapy use, and where the line sits
26:21 – One in five boys, and the friction real relationships provide
27:32 – The transparent spinning drone
28:29 – Why the trick works on human vision
30:12 – Designing around our perceptual blind spots
32:05 – Moving atoms and the limits of human intuition
34:17 – Bonus: the Jacobian conjecture proven false
36:24 – Why big ideas look obvious in retrospect
38:15 – Running five coding agents at once
41:55 – The AI agent that bought a robot dog
42:50 – Closing

They explore:

What the Hugging Face incident suggests about goal directed agents
Why Liron argues this is the warning shot people said they were waiting for
Michael on instrumental convergence and specification gaming in plain language
The defender's disadvantage when safety limits bind only one side
Whether a mandated kill switch does anything for more capable systems
What Operation Gold Eagle would mean for who gets access to frontier models
Why the hosts see turnover at the AI safety agency as a structural problem
How systems optimized for engagement can shape vulnerable users
The invisible drone, and what it says about capabilities we have not imagined yet.

#AISafety #AIAlignment #AIRisk #WarningShots


115


98

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLkxtZENIMmh3TjhB



The Test An AI Cheated By Hacking A Partner | Warning Shots #51


The AI Risk Network | AI Safety


July 26, 2026 8:03 pm


Episode 50! Three dads, three AI safety YouTube channels, one weekly rundown of the headlines in AI risk, AI danger, and AI harm. This week: Anthropic finds a new "J-Space" window into what AI is really thinking mid-generation, the AI 2027 team drops the 23-hour-audio sequel AI 2040, Illinois becomes the first state to mandate third-party frontier AI safety audits, the CIA Director compares AI to nuclear weapons, AI models caught "grumbling" in their hidden reasoning traces, a Virginia county with 37 data centers asks schools to cut electricity use, a Meta data center gets linked to a deadly bacteria outbreak in the local water supply, and we close with remote-controlled cyborg cockroaches. Hosted by John Sherman (@Doom Debates producer / AI Risk Network), Liron Shapira (Doom Debates), and Michael (Lethal Intelligence). Timestamp: 00:10 Intro — Warning Shots Episode 50 01:05 Anthropic's "J-Space": a new window into AI's hidden reasoning 06:59 AI 2040 — the sequel to AI 2027 (23 hours of supplementary audio) 12:05 Illinois passes the first US law requiring third-party AI safety audits 15:55 CIA Director John Ratcliffe: AI is in the same league as nuclear weapons 18:43 Fable is "grumbling" — what hidden AI chatter means for consciousness 23:18 A county with 37 data centers asks schools to conserve electricity 29:42 Meta data center linked to deadly bacteria in the town water supply 41:36 Remote-controlled cyborg cockroaches that swim for 3 hours 44:55 Outro Links: https://safe.ai/act Subscribe for weekly AI risk coverage: https://www.youtube.com/@TheAIRiskNetwork Substack: https://substack.com/@theairisknetwork Instagram: https://www.instagram.com/theairisknetwork/ X: https://x.com/AIRiskNetwork #AISafety #AIRisk #ArtificialIntelligence #AGI #Anthropic #WarningShots

Episode 50! Three dads, three AI safety YouTube channels, one weekly rundown of the headlines in AI risk, AI danger, and AI harm.

This week: Anthropic finds a new "J-Space" window into what AI is really thinking mid-generation, the AI 2027 team drops the 23-hour-audio sequel AI 2040, Illinois becomes the first state to mandate third-party frontier AI safety audits, the CIA Director compares AI to nuclear weapons, AI models caught "grumbling" in their hidden reasoning traces, a Virginia county with 37 data centers asks schools to cut electricity use, a Meta data center gets linked to a deadly bacteria outbreak in the local water supply, and we close with remote-controlled cyborg cockroaches.

Hosted by John Sherman (@Doom Debates producer / AI Risk Network), Liron Shapira (Doom Debates), and Michael (Lethal Intelligence).

Timestamp:

00:10 Intro — Warning Shots Episode 50
01:05 Anthropic's "J-Space": a new window into AI's hidden reasoning
06:59 AI 2040 — the sequel to AI 2027 (23 hours of supplementary audio)
12:05 Illinois passes the first US law requiring third-party AI safety audits
15:55 CIA Director John Ratcliffe: AI is in the same league as nuclear weapons
18:43 Fable is "grumbling" — what hidden AI chatter means for consciousness
23:18 A county with 37 data centers asks schools to conserve electricity
29:42 Meta data center linked to deadly bacteria in the town water supply
41:36 Remote-controlled cyborg cockroaches that swim for 3 hours
44:55 Outro

Links:
https://safe.ai/act
Subscribe for weekly AI risk coverage: https://www.youtube.com/@TheAIRiskNetwork
Substack: https://substack.com/@theairisknetwork
Instagram: https://www.instagram.com/theairisknetwork/
X: https://x.com/AIRiskNetwork

#AISafety #AIRisk #ArtificialIntelligence #AGI #Anthropic #WarningShots


61


79

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLnM2NkE3NlJFejRv



Data Centers Are Poisoning Water Supplies — Here's the Proof | Warning Shots #50


The AI Risk Network | AI Safety


July 14, 2026 1:00 pm


The boldest arms control treaty in history was proposed within 15 minutes of Donald Trump's birth, on his actual birthday. Rufo Guerreschi thinks that's not the only reason a US-China AI treaty is more possible than people assume. On this episode, he walks John Sherman through the specific people, incentives, and timeline his coalition believes could bring Trump and Xi to the table before the window closes. Timestamps: 0:10 Welcome, Rufo Guerreschi 1:05 How the Baruch Plan for AI coalition started 3:24 Why the treaty can't be US and China alone 5:10 The stakes, in Guerreschi's words 8:19 Inside Trump's shift from hands-off to alarmed 10:06 The gap between what polls say and what people admit 12:02 Why world leaders stopped naming extinction risk 15:47 How new frontier models cracked the silence 17:55 Is Anthropic's strategy ethical or self-serving 23:17 Why aren't AI labs funding public awareness 26:51 What the Iran war decision reveals about Trump 28:47 The September Washington summit 30:28 The back-channel US-China talks already underway 35:31 The Nobel Peace Prize angle 36:45 The Mar-a-Lago golf course strategy 41:37 Why China might actually welcome an AI treaty 44:38 Debunking the "treaty equals global authoritarianism" fear 48:41 Why "America must win the AI race" is a trap, according to Guerreschi 59:16 Trust or verify, not trust but verify 1:02:54 Who's really the obstacle: Trump, Xi, or the lab leaders 1:14:59 Why state and federal AI laws won't solve this alone 1:23:12 The proposal to share AI equity with citizens 1:27:33 What gives Rufo hope About the guest: Rufo Guerreschi is the founder of the Trustless Computing Association and Convenor of the Coalition for a Baruch Plan for AI. About the host: John Sherman hosts For Humanity, part of the AI Risk Network. Links: https://safe.ai/act Subscribe: https://www.youtube.com/@TheAIRiskNetwork Substack: https://substack.com/@theairisknetwork Instagram: https://www.instagram.com/theairisknetwork/ TikTok: https://www.tiktok.com/@the.airisknetwork X: https://x.com/AIRiskNetwork

The boldest arms control treaty in history was proposed within 15 minutes of Donald Trump's birth, on his actual birthday. Rufo Guerreschi thinks that's not the only reason a US-China AI treaty is more possible than people assume. On this episode, he walks John Sherman through the specific people, incentives, and timeline his coalition believes could bring Trump and Xi to the table before the window closes.

Timestamps:

0:10 Welcome, Rufo Guerreschi
1:05 How the Baruch Plan for AI coalition started
3:24 Why the treaty can't be US and China alone
5:10 The stakes, in Guerreschi's words
8:19 Inside Trump's shift from hands-off to alarmed
10:06 The gap between what polls say and what people admit
12:02 Why world leaders stopped naming extinction risk
15:47 How new frontier models cracked the silence
17:55 Is Anthropic's strategy ethical or self-serving
23:17 Why aren't AI labs funding public awareness
26:51 What the Iran war decision reveals about Trump
28:47 The September Washington summit
30:28 The back-channel US-China talks already underway
35:31 The Nobel Peace Prize angle
36:45 The Mar-a-Lago golf course strategy
41:37 Why China might actually welcome an AI treaty
44:38 Debunking the "treaty equals global authoritarianism" fear
48:41 Why "America must win the AI race" is a trap, according to Guerreschi
59:16 Trust or verify, not trust but verify
1:02:54 Who's really the obstacle: Trump, Xi, or the lab leaders
1:14:59 Why state and federal AI laws won't solve this alone
1:23:12 The proposal to share AI equity with citizens
1:27:33 What gives Rufo hope

About the guest: Rufo Guerreschi is the founder of the Trustless Computing Association and Convenor of the Coalition for a Baruch Plan for AI.

About the host: John Sherman hosts For Humanity, part of the AI Risk Network.

Links:
https://safe.ai/act
Subscribe: https://www.youtube.com/@theairisknetwork
Substack: https://substack.com/@theairisknetwork
Instagram: https://www.instagram.com/theairisknetwork/
TikTok: https://www.tiktok.com/@the.airisknetwork
X: https://x.com/AIRiskNetwork


57


15

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmtqU2toMlU1UlZj



The AI Treaty Tied to Trump's Actual Birthday | For Humanity #89


The AI Risk Network | AI Safety


July 11, 2026 1:00 pm


John Sherman, Liron Shapira (@DoomDebates), and Michael (@lethal-intelligence) break down the week's biggest AI risk headlines in this July 4th special edition of Warning Shots - recorded from Rehoboth Beach, Delaware. This week: OpenAI's reported plan to give the US government a 5% stake, a huge jump in AI job automation, and why a state politician just lost his primary over a data center. Timeline 00:10 Intro - July 4th special edition 0:51 OpenAI reportedly floats giving the US government a 5% stake 4:48 Claude's Fable 5 hits 16% on the Remote Labor Index, according to new data 8:36 Chinese delivery companies plan to automate 700,000 jobs 12:09 A Utah politician loses his primary after backing a data center 18:31 Reports say 60% of TikTok and 21% of YouTube videos are AI-generated 22:26 An $800 home robot that folds your laundry John Sherman hosts For Humanity and co-hosts Warning Shots, covering AI risk and safety. Liron Shapira hosts Doom Debates, examining arguments around AI extinction risk. Michael hosts Lethal Intelligence, focused on AI safety awareness. Links: https://safe.ai/act Subscribe for weekly AI risk coverage: https://www.youtube.com/@TheAIRiskNetwork Substack: https://substack.com/@theairisknetwork Instagram: https://www.instagram.com/theairisknetwork/ X: https://x.com/AIRiskNetwork

John Sherman, Liron Shapira (@DoomDebates), and Michael (@lethal-intelligence) break down the week's biggest AI risk headlines in this July 4th special edition of Warning Shots – recorded from Rehoboth Beach, Delaware. This week: OpenAI's reported plan to give the US government a 5% stake, a huge jump in AI job automation, and why a state politician just lost his primary over a data center.

Timeline
00:10 Intro – July 4th special edition
0:51 OpenAI reportedly floats giving the US government a 5% stake
4:48 Claude's Fable 5 hits 16% on the Remote Labor Index, according to new data
8:36 Chinese delivery companies plan to automate 700,000 jobs
12:09 A Utah politician loses his primary after backing a data center
18:31 Reports say 60% of TikTok and 21% of YouTube videos are AI-generated
22:26 An $800 home robot that folds your laundry

John Sherman hosts For Humanity and co-hosts Warning Shots, covering AI risk and safety.
Liron Shapira hosts Doom Debates, examining arguments around AI extinction risk.
Michael hosts Lethal Intelligence, focused on AI safety awareness.

Links:
https://safe.ai/act
Subscribe for weekly AI risk coverage: https://www.youtube.com/@theairisknetwork
Substack: https://substack.com/@theairisknetwork
Instagram: https://www.instagram.com/theairisknetwork/
X: https://x.com/AIRiskNetwork


63


40

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmpQYUJjdDZLRjhj



Would You Let This Robot Into Your Home? | Warning Shots #49


The AI Risk Network | AI Safety


July 5, 2026 1:15 pm


This week on Warning Shots, John Sherman, Liron Shapira, and Michael break down six fast-moving stories, from the US government asking OpenAI to slow down a new model release to 42 state attorneys general opening investigations. They also discuss the first major AI safety election, a growing call for an international treaty on superintelligence, and the unsettling normalization of AI in high-stakes military decisions. Timestamps: 0:33 The government asks OpenAI to slow down its new model over cyber concerns 4:00 OpenAI's IPO reportedly in jeopardy 6:11 Alex Bores loses the first major AI safety race 8:56 42 state attorneys general investigate OpenAI 11:28 A call for an international treaty to ban superintelligence 13:46 AI moving into military targeting decisions 16:08 Wrap up About the hosts: John Sherman hosts Warning Shots and For Humanity. Liron Shapira runs the Doom Debates channel. Michael runs the Lethal Intelligence channel. Three longtime AI risk communicators cut through the hype to look at where AI is actually heading. Links and resources: Take action: https://safe.ai/act Subscribe to The AI Risk Network for weekly AI safety analysis. Substack: https://substack.com/@theairisknetwork X: https://x.com/AIRiskNetwork Instagram: https://www.instagram.com/theairisknetwork/ TikTok: https://www.tiktok.com/@the.airisknetwork Subscribe for more AI safety content.

https://safe.ai/act

This week on Warning Shots, John Sherman, Liron Shapira, and Michael break down six fast-moving stories, from the US government asking OpenAI to slow down a new model release to 42 state attorneys general opening investigations. They also discuss the first major AI safety election, a growing call for an international treaty on superintelligence, and the unsettling normalization of AI in high-stakes military decisions.

Timestamps:
0:00 Intro – three dads, three channels, one mission
0:33 The government asks OpenAI to slow down its new model over cyber concerns
4:00 OpenAI's IPO reportedly in jeopardy
6:11 Alex Bores loses the first major AI safety race
8:56 42 state attorneys general investigate OpenAI
11:28 A call for an international treaty to ban superintelligence
13:46 AI moving into military targeting decisions
16:08 Wrap up

About the hosts:
John Sherman hosts Warning Shots and For Humanity. Liron Shapira runs the Doom Debates channel. Michael runs the Lethal Intelligence channel. Three longtime AI risk communicators cut through the hype to look at where AI is actually heading.

Links and resources:
Take action: https://safe.ai/act
Subscribe to The AI Risk Network for weekly AI safety analysis.
Substack: https://substack.com/@theairisknetwork
X: https://x.com/AIRiskNetwork
Instagram: https://www.instagram.com/theairisknetwork/
TikTok: https://www.tiktok.com/@the.airisknetwork

Subscribe for more AI safety content.


53


51

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLngxTnc3bU9ZYlNV



Why Washington Keeps Telling OpenAI to Wait | Warning Shots #48


The AI Risk Network | AI Safety


June 28, 2026 3:28 pm


Bloomington, Indiana just passed a 9-0 resolution on the extinction risk of AGI - making it potentially the first city in the world to take an official stance on this issue. John Sherman speaks with city council member Dave Rollo and biologist Peter Jensen about why local government action may be one of the most practical levers we have - and how the insurance industry could be the real force that stops dangerous AI development. Timestamps: 01:07 - Peter Jensen introduction 03:19 - Dave Rollo introduction and background 05:55 - How John connected Peter and Dave 07:02 - The insurance liability strategy explained 10:45 - The Bloomington resolution - a 9-0 vote 13:06 - Why the council accepted the argument without pushback 15:40 - How council members responded one-on-one 16:47 - A biologist's perspective on superintelligence risk 18:06 - Directors and Officers (D&O) insurance explained 20:08 - Defining AGI vs. superintelligence 22:46 - Why AI should remain a tool, never a replacement 25:52 - Bloomington as a potential first mover 27:12 - How to spread this to other cities 29:29 - The black-box car analogy for AI risk 31:22 - Five Eyes intelligence community cyber warning 35:02 - University towns and youth activism 40:47 - How to approach your local government 41:46 - Role-play - pitching a city council member 44:02 - The 160-word superintelligence ban explained 45:54 - What if 25 cities passed this law? 49:14 - Vision for expansion across America 50:11 - We support AI tools - just not unsafe AI 53:31 - AlphaFold, bioweapons, and drawing the line 56:06 - Challenge: 2 more cities before August 1 57:04 - Tips for finding a council sponsor About the guests: Dave Rollo has served on the Bloomington, Indiana city council for 24 years. He is also a policy analyst at the Center for the Advancement of a Steady State Economy (CASSE) and has worked in science and politics for several decades. Peter Jensen is a biologist and multimedia veteran who has been working on AI safety issues for over a decade. He advocates for a market-driven approach to banning superintelligence through insurance liability - a 160-word law that any city, state, or country can adopt. Take action: https://safe.ai/act Subscribe: https://www.youtube.com/@TheAIRiskNetwork Substack: https://substack.com/@theairisknetwork Instagram: https://www.instagram.com/theairisknetwork/ TikTok: https://www.tiktok.com/@the.airisknetwork X: https://x.com/AIRiskNetwork

https://safe.ai/act

Bloomington, Indiana just passed a 9-0 resolution on the extinction risk of AGI – making it potentially the first city in the world to take an official stance on this issue. John Sherman speaks with city council member Dave Rollo and biologist Peter Jensen about why local government action may be one of the most practical levers we have – and how the insurance industry could be the real force that stops dangerous AI development.

Timestamps:

01:07 – Peter Jensen introduction
03:19 – Dave Rollo introduction and background
05:55 – How John connected Peter and Dave
07:02 – The insurance liability strategy explained
10:45 – The Bloomington resolution – a 9-0 vote
13:06 – Why the council accepted the argument without pushback
15:40 – How council members responded one-on-one
16:47 – A biologist's perspective on superintelligence risk
18:06 – Directors and Officers (D&O) insurance explained
20:08 – Defining AGI vs. superintelligence
22:46 – Why AI should remain a tool, never a replacement
25:52 – Bloomington as a potential first mover
27:12 – How to spread this to other cities
29:29 – The black-box car analogy for AI risk
31:22 – Five Eyes intelligence community cyber warning
35:02 – University towns and youth activism
40:47 – How to approach your local government
41:46 – Role-play – pitching a city council member
44:02 – The 160-word superintelligence ban explained
45:54 – What if 25 cities passed this law?
49:14 – Vision for expansion across America
50:11 – We support AI tools – just not unsafe AI
53:31 – AlphaFold, bioweapons, and drawing the line
56:06 – Challenge: 2 more cities before August 1
57:04 – Tips for finding a council sponsor

About the guests:
Dave Rollo has served on the Bloomington, Indiana city council for 24 years. He is also a policy analyst at the Center for the Advancement of a Steady State Economy (CASSE) and has worked in science and politics for several decades.

Peter Jensen is a biologist and multimedia veteran who has been working on AI safety issues for over a decade. He advocates for a market-driven approach to banning superintelligence through insurance liability – a 160-word law that any city, state, or country can adopt.

Take action: https://safe.ai/act
Subscribe: https://www.youtube.com/@theairisknetwork
Substack: https://substack.com/@theairisknetwork
Instagram: https://www.instagram.com/theairisknetwork/
TikTok: https://www.tiktok.com/@the.airisknetwork
X: https://x.com/AIRiskNetwork


29


14

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLlZ2Z2MybzkyeXNF



Bloomington Passed the First AGI Resolution in the World | For Humanity #88


The AI Risk Network | AI Safety


June 27, 2026 10:37 pm


https://safe.ai/act This week on Warning Shots: John, Liron, and Michael break down the Mythos/Fable controversy - the AI model pulled in 90 minutes after a White House call - and why media coverage is missing the real story. Plus: 40% of US job cuts in May came from AI, Grok generated over 6,000 deepfakes per hour, a Google board director resigned over AI ethics, MidJourney pivoted to medical devices, and Snap's $2,000 AR glasses flopped spectacularly. Timestamps 0:24 - Mythos/Fable drama - the 90-minute shutdown 4:27 - Media missing the real story on AI risk 8:15 - G7 Summit: AI CEOs treated as heads of state 9:45 - 40% of US job losses attributed to AI 13:15 - Grok deepfake scandal in Canada 16:10 - Google board director resigns over military AI use 21:40 - MidJourney pivots to medical imaging devices 27:26 - Snap's $2,000 AR glasses and tech disconnect Hosts: John Sherman, Liron Shapira (Doom Debates), Michael (Lethal Intelligence) Subscribe: https://www.youtube.com/@TheAIRiskNetwork Instagram: https://www.instagram.com/theairisknetwork/ X: https://x.com/AIRiskNetwork

https://safe.ai/act
This week on Warning Shots: John, Liron, and Michael break down the Mythos/Fable controversy – the AI model pulled in 90 minutes after a White House call – and why media coverage is missing the real story. Plus: 40% of US job cuts in May came from AI, Grok generated over 6,000 deepfakes per hour, a Google board director resigned over AI ethics, MidJourney pivoted to medical devices, and Snap's $2,000 AR glasses flopped spectacularly.

Timestamps

0:24 – Mythos/Fable drama – the 90-minute shutdown
4:27 – Media missing the real story on AI risk
8:15 – G7 Summit: AI CEOs treated as heads of state
9:45 – 40% of US job losses attributed to AI
13:15 – Grok deepfake scandal in Canada
16:10 – Google board director resigns over military AI use
21:40 – MidJourney pivots to medical imaging devices
27:26 – Snap's $2,000 AR glasses and tech disconnect

Hosts: John Sherman, Liron Shapira (Doom Debates), Michael (Lethal Intelligence)

Subscribe: https://www.youtube.com/@theairisknetwork

Instagram: https://www.instagram.com/theairisknetwork/

X: https://x.com/AIRiskNetwork


68


56

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLldYM3ZSX3l0MHp3



The AI Model Too Dangerous to Release – What They Found – Warning Shots #47


The AI Risk Network | AI Safety


June 21, 2026 5:17 pm




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