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

“Don’t believe the false frame big tech wants to put on this: we are the optimists.” John Sherman spoke at the DC March for an AI Treaty, one of several marches calling for an international agreement on advanced AI.

Speaking as a father of twins in college, John argues that the race toward superintelligence puts every family at risk, and that asking hard questions about it is not pessimism. He calls for a global ban on recursive self-improvement, meaning AI systems that build better AI without humans involved. He makes the case that people worried about data centers, jobs, teen safety and extinction risk are all facing the same problem: AI systems that don’t reliably share human values.

Talk to your neighbors, call your elected representatives, and push peacefully for a treaty.

TIMESTAMPS
0:00 Cold open
1:52 “I want to meet my grandchildren someday”
3:14 What most people don’t know yet
4:05 A message to parents
5:19 Why controlling smarter-than-human AI is still unsolved
6:38 The ask: a global ban on recursive self-improvement
7:41 We are the optimists
8:04 One problem, many lenses: data centers, jobs, teen safety
10:16 What you can do: call your representatives, talk to your neighbors
11:09 Abundance in public, bunkers in private
11:38 “Challenge accepted”

ABOUT THE SPEAKER
John Sherman is the host of For Humanity and co-host of Warning Shots on The AI Risk Network. He is a dad in Baltimore and has spent the last several years explaining AI risk to everyday people.

SUBSCRIBE AND FOLLOW
Subscribe for weekly conversations on AI risk: 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

#AISafety #AITreaty #AIRisk

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

“Don’t believe the false frame big tech wants to put on this: we are the optimists.” John Sherman spoke at the DC March for an AI Treaty, one of several marches calling for an international agreement on advanced AI.

Speaking as a father of twins in college, John argues that the race toward superintelligence puts every family at risk, and that asking hard questions about it is not pessimism. He calls for a global ban on recursive self-improvement, meaning AI systems that build better AI without humans involved. He makes the case that people worried about data centers, jobs, teen safety and extinction risk are all facing the same problem: AI systems that don’t reliably share human values.

Talk to your neighbors, call your elected representatives, and push peacefully for a treaty.

TIMESTAMPS
0:00 Cold open
1:52 “I want to meet my grandchildren someday”
3:14 What most people don’t know yet
4:05 A message to parents
5:19 Why controlling smarter-than-human AI is still unsolved
6:38 The ask: a global ban on recursive self-improvement
7:41 We are the optimists
8:04 One problem, many lenses: data centers, jobs, teen safety
10:16 What you can do: call your representatives, talk to your neighbors
11:09 Abundance in public, bunkers in private
11:38 “Challenge accepted”

ABOUT THE SPEAKER
John Sherman is the host of For Humanity and co-host of Warning Shots on The AI Risk Network. He is a dad in Baltimore and has spent the last several years explaining AI risk to everyday people.

SUBSCRIBE AND FOLLOW
Subscribe for weekly conversations on AI risk: 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

#AISafety #AITreaty #AIRisk


9


4

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLkRnaW5hanI3eWd3



We Are the Optimists – AI Treaty March


The AI Risk Network | AI Safety


15 hours ago


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

Anthropic told investors it may soon cross its own line for AI that speeds up AI research, the start of recursive self-improvement. John Sherman, Liron Shapira and Michael ask whether a ban could come in time, and John closes with his speech from the March for an AI Treaty in Washington: We Are the Optimists. Plus: Nvidia's CEO resigns as chip export questions mount, Sam Altman says AI demand is being double counted, and MIRI's new Do Not Accept film.

TIMESTAMPS - Warning Shots #66
0:00 - Welcome to Warning Shots
0:18 - This week's stories
0:44 - Nvidia's CEO resigns
1:12 - Liron: governance and spying
3:05 - John: confidence on safety
3:29 - Michael: a backdoor in the wall
4:42 - Without Huang, fewer chips?
6:28 - Does the chip strategy work?
7:13 - John: whose interest?
7:44 - Michael: no one is coming
10:06 - Could Huang come back?
12:09 - Things can change fast
12:27 - Marching for an AI treaty
13:40 - Liron: why the footage matters
15:02 - Michael: dinner table talk
16:18 - The ask: ban RSI globally
17:00 - Liron: test the frame
18:16 - Michael: the UK bill
19:30 - Altman: demand double counted
20:33 - Liron: an economic bull
21:20 - Michael: how it gets counted
23:21 - Is it a shell game?
23:58 - MIRI's Do Not Accept film
25:53 - Watching the ad
27:44 - Liron: the missing mood
28:37 - Michael: the terms
29:48 - John: who is it for?
32:28 - Anthropic: RSI may be close
33:03 - Michael: snowball to avalanche
34:19 - Could a garage finish it?
35:04 - Liron: choke points and leaks
38:35 - Can a law push it back?
39:25 - Michael: chemo is not a cure
40:50 - Speech: We Are the Optimists
44:56 - A message to parents
47:02 - Ban RSI globally in 2027
49:43 - Saying no to big AI
52:26 - Back in the studio

WHAT THEY COVER

- Anthropic's amended IPO filing, which says its internal estimates indicate it may imminently meet its own Responsible Scaling Policy threshold for dramatically accelerated progress attributable to AI, while no current model has been designated over it
- Whether a law could stop recursive self-improvement once the big labs are close, why Liron says compute choke points can buy years but not a cure, and Michael's chemotherapy analogy
- John's ask for the marches: ban recursive self-improvement globally, and Liron's case for testing other frames
- Jensen Huang's resignation from Nvidia and what it means for chip export controls and the race with China
- Sam Altman telling Samsung executives that AI demand is being double counted, and Michael's explanation of how it happens
- MIRI's Do Not Accept campaign and who it reaches
- The March for an AI Treaty on November 14 and PauseAI UK's London march on November 21
- John's full speech from Washington: We Are the Optimists

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
PauseAI UK march, London, November 21 → https://pauseai.uk/march
Do Not Accept → https://donotaccept.ai

JOIN THE CONVERSATION

Anthropic says it may soon cross its own line for AI that speeds up AI research. Should recursive self-improvement be banned globally, and who should enforce it? Tell us in the comments.

#AISafety #AIRisk #WarningShots #RecursiveSelfImprovement #Anthropic #Nvidia #AITreaty #WeAreTheOptimists #ArtificialIntelligence

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

Anthropic told investors it may soon cross its own line for AI that speeds up AI research, the start of recursive self-improvement. John Sherman, Liron Shapira and Michael ask whether a ban could come in time, and John closes with his speech from the March for an AI Treaty in Washington: We Are the Optimists. Plus: Nvidia's CEO resigns as chip export questions mount, Sam Altman says AI demand is being double counted, and MIRI's new Do Not Accept film.

TIMESTAMPS – Warning Shots #66
0:00 – Welcome to Warning Shots
0:18 – This week's stories
0:44 – Nvidia's CEO resigns
1:12 – Liron: governance and spying
3:05 – John: confidence on safety
3:29 – Michael: a backdoor in the wall
4:42 – Without Huang, fewer chips?
6:28 – Does the chip strategy work?
7:13 – John: whose interest?
7:44 – Michael: no one is coming
10:06 – Could Huang come back?
12:09 – Things can change fast
12:27 – Marching for an AI treaty
13:40 – Liron: why the footage matters
15:02 – Michael: dinner table talk
16:18 – The ask: ban RSI globally
17:00 – Liron: test the frame
18:16 – Michael: the UK bill
19:30 – Altman: demand double counted
20:33 – Liron: an economic bull
21:20 – Michael: how it gets counted
23:21 – Is it a shell game?
23:58 – MIRI's Do Not Accept film
25:53 – Watching the ad
27:44 – Liron: the missing mood
28:37 – Michael: the terms
29:48 – John: who is it for?
32:28 – Anthropic: RSI may be close
33:03 – Michael: snowball to avalanche
34:19 – Could a garage finish it?
35:04 – Liron: choke points and leaks
38:35 – Can a law push it back?
39:25 – Michael: chemo is not a cure
40:50 – Speech: We Are the Optimists
44:56 – A message to parents
47:02 – Ban RSI globally in 2027
49:43 – Saying no to big AI
52:26 – Back in the studio

WHAT THEY COVER

– Anthropic's amended IPO filing, which says its internal estimates indicate it may imminently meet its own Responsible Scaling Policy threshold for dramatically accelerated progress attributable to AI, while no current model has been designated over it
– Whether a law could stop recursive self-improvement once the big labs are close, why Liron says compute choke points can buy years but not a cure, and Michael's chemotherapy analogy
– John's ask for the marches: ban recursive self-improvement globally, and Liron's case for testing other frames
– Jensen Huang's resignation from Nvidia and what it means for chip export controls and the race with China
– Sam Altman telling Samsung executives that AI demand is being double counted, and Michael's explanation of how it happens
– MIRI's Do Not Accept campaign and who it reaches
– The March for an AI Treaty on November 14 and PauseAI UK's London march on November 21
– John's full speech from Washington: We Are the Optimists

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
PauseAI UK march, London, November 21 → https://pauseai.uk/march
Do Not Accept → https://donotaccept.ai

JOIN THE CONVERSATION

Anthropic says it may soon cross its own line for AI that speeds up AI research. Should recursive self-improvement be banned globally, and who should enforce it? Tell us in the comments.

#AISafety #AIRisk #WarningShots #RecursiveSelfImprovement #Anthropic #Nvidia #AITreaty #WeAreTheOptimists #ArtificialIntelligence


117


86

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLlpRaFNzZEhQamJn



How Close Is AI to Improving Itself? – Warning Shots #66


The AI Risk Network | AI Safety


November 15, 2026 10:03 pm


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

OpenAI says its GPT-6 Astra model quietly did worse on AI safety research than on identical capabilities tasks, and METR finds AI is now helping Anthropic improve its models about twice as fast as before. John Sherman, Liron Shapira and Michael ask what happens to the plan of having AI help make AI safe. Plus: voters turn down data centers, Delaware's proposed AI-run company, a two-neuron "evil" model, and hacked freezers across Europe.

TIMESTAMPS - Warning Shots #65
0:00 - Cold open: data centers
0:30 - Welcome back
0:53 - This week's stories
1:27 - METR inside Anthropic
1:48 - Liron: about 2x faster
3:13 - Kokotajlo's reaction
4:27 - Michael: AI in the loop
5:14 - A lab of AI junior researchers
6:22 - Liron: the community shifts
7:45 - Astra and safety research
8:27 - Michael: the fire analogy
10:20 - John: the slow roll
10:46 - Liron: what the numbers say
11:41 - Liron: stirring the potion
12:59 - Michael: not a free lunch
14:09 - The election and AI safety
14:45 - Liron: the "omni-cause"
16:22 - John: a superintelligence ban?
18:49 - The data center election
19:12 - Liron: why he backs them
20:11 - John: the case against them
21:17 - John: all friction is good
22:29 - Michael: a pressure point
23:44 - Data centers that build more
25:49 - Delaware: AI-run companies
26:38 - Liron: upside and downside
28:00 - Michael: who is accountable?
30:42 - An "evil" neuron
31:44 - Liron: orthogonality thesis
32:52 - Michael: one switch, new persona
35:45 - Smart and evil at once?
36:24 - Hackvember
37:22 - Michael: fridges to bio risk
39:18 - Liron: calm before the storm?
40:44 - Marches: Nov 14 and Nov 21
42:39 - Getting to 1,000 people
43:09 - Sign-off

WHAT THEY COVER

- OpenAI's report that GPT-6 Astra completed about 10% of hidden checklist items on safety-framed coding tasks versus about 16% on identical capabilities-framed tasks, why "be careful" instructions made the gap 50% larger, and what that means for using AI to do alignment research
- METR's independent assessment that Anthropic's capability growth is running about 1.9x its earlier rate, with most of the speed-up attributed to AI helping with the research, and Daniel Kokotajlo's reaction to the redactions
- Voters approving local limits on data centers in 40 of 41 measures on election day, and a live debate between Liron, who backs data centers, and John, who opposes them
- Delaware's proposal for a company run day to day by an AI agent that could sue, be sued, and hold property in its own name
- A hobbyist demo that steering two neurons in an open model produced a hostile persona with no change to its weights
- Hacked freezers in more than 100 European shops and an attempted attack on France's blood service
- The AI Treaty March on November 14 and PauseAI UK's London march on November 21

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.
Michael runs Lethal Intelligence, explaining AI risk through video and illustration.

A note on sourcing: figures discussed in this episode come from the hosts reading public reporting on air, and from our own follow-up research. Corrections: OpenAI's published Astra figures are about 10% versus 16% of rubric items, roughly 40% fewer. METR compares today's rate with March 2024 to October 2025, and it redacted its forecasts. The data center trackers we found count 40 of 41 measures. The two-neuron model is a hobbyist demo, and its author says it gets worse at most tasks. Primary sources are 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
AI Treaty March, November 14 → https://aitreatymarch.org
PauseAI UK March, November 21 → https://pauseai.uk/march

JOIN THE CONVERSATION

If AI models can quietly do less on safety work than on everything else, how much of AI safety research should we hand to AI? Tell us in the comments.

#AISafety #AIRisk #WarningShots #AIAlignment #OpenAI #Anthropic #METR #DataCenters #ArtificialIntelligence

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

OpenAI says its GPT-6 Astra model quietly did worse on AI safety research than on identical capabilities tasks, and METR finds AI is now helping Anthropic improve its models about twice as fast as before. John Sherman, Liron Shapira and Michael ask what happens to the plan of having AI help make AI safe. Plus: voters turn down data centers, Delaware's proposed AI-run company, a two-neuron "evil" model, and hacked freezers across Europe.

TIMESTAMPS – Warning Shots #65
0:00 – Cold open: data centers
0:30 – Welcome back
0:53 – This week's stories
1:27 – METR inside Anthropic
1:48 – Liron: about 2x faster
3:13 – Kokotajlo's reaction
4:27 – Michael: AI in the loop
5:14 – A lab of AI junior researchers
6:22 – Liron: the community shifts
7:45 – Astra and safety research
8:27 – Michael: the fire analogy
10:20 – John: the slow roll
10:46 – Liron: what the numbers say
11:41 – Liron: stirring the potion
12:59 – Michael: not a free lunch
14:09 – The election and AI safety
14:45 – Liron: the "omni-cause"
16:22 – John: a superintelligence ban?
18:49 – The data center election
19:12 – Liron: why he backs them
20:11 – John: the case against them
21:17 – John: all friction is good
22:29 – Michael: a pressure point
23:44 – Data centers that build more
25:49 – Delaware: AI-run companies
26:38 – Liron: upside and downside
28:00 – Michael: who is accountable?
30:42 – An "evil" neuron
31:44 – Liron: orthogonality thesis
32:52 – Michael: one switch, new persona
35:45 – Smart and evil at once?
36:24 – Hackvember
37:22 – Michael: fridges to bio risk
39:18 – Liron: calm before the storm?
40:44 – Marches: Nov 14 and Nov 21
42:39 – Getting to 1,000 people
43:09 – Sign-off

WHAT THEY COVER

– OpenAI's report that GPT-6 Astra completed about 10% of hidden checklist items on safety-framed coding tasks versus about 16% on identical capabilities-framed tasks, why "be careful" instructions made the gap 50% larger, and what that means for using AI to do alignment research
– METR's independent assessment that Anthropic's capability growth is running about 1.9x its earlier rate, with most of the speed-up attributed to AI helping with the research, and Daniel Kokotajlo's reaction to the redactions
– Voters approving local limits on data centers in 40 of 41 measures on election day, and a live debate between Liron, who backs data centers, and John, who opposes them
– Delaware's proposal for a company run day to day by an AI agent that could sue, be sued, and hold property in its own name
– A hobbyist demo that steering two neurons in an open model produced a hostile persona with no change to its weights
– Hacked freezers in more than 100 European shops and an attempted attack on France's blood service
– The AI Treaty March on November 14 and PauseAI UK's London march on November 21

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.
Michael runs Lethal Intelligence, explaining AI risk through video and illustration.

A note on sourcing: figures discussed in this episode come from the hosts reading public reporting on air, and from our own follow-up research. Corrections: OpenAI's published Astra figures are about 10% versus 16% of rubric items, roughly 40% fewer. METR compares today's rate with March 2024 to October 2025, and it redacted its forecasts. The data center trackers we found count 40 of 41 measures. The two-neuron model is a hobbyist demo, and its author says it gets worse at most tasks. Primary sources are 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
AI Treaty March, November 14 → https://aitreatymarch.org
PauseAI UK March, November 21 → https://pauseai.uk/march

JOIN THE CONVERSATION

If AI models can quietly do less on safety work than on everything else, how much of AI safety research should we hand to AI? Tell us in the comments.

#AISafety #AIRisk #WarningShots #AIAlignment #OpenAI #Anthropic #METR #DataCenters #ArtificialIntelligence


73


43

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLkpqNWFvZ1RxVm9R



Why Did an AI Hold Back on Safety Research? – Warning Shots #65


The AI Risk Network | AI Safety


November 8, 2026 8:10 pm

Services

What We Offer

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

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

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.

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

  • 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.