For Humanity Podcast

0

0

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLjlYWGdWczBlQXBz

AI autonomously solves Millennium Math Prize problem

The AI Risk Network | AI Safety

97 minutes ago

4

1

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLnVVVzJjdmFIb1dN

SoftBank CEO predicts 100 trillion AIS

The AI Risk Network | AI Safety

21 hours ago

1

0

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLlRGcFR4NkUtd2Z3

Would people turn off AI if they could?

The AI Risk Network | AI Safety

September 16, 2026 7:15 pm

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

Two days after an Anthropic researcher's resignation post went viral, Axios co-founder Jim VandeHei joins John Sherman to talk about what changed in Washington, and what he says still has not been built. VandeHei explains why he doubts this week's burst of congressional attention turns into real legislation, describes the independent review board and "kill switch" he thinks the industry still lacks, and tells John why his own odds of AI causing serious harm are lower than his host's.

They also get into VandeHei's new book "Simplify," what AI has done for his family, AI in schools, and a moment from an advertising industry panel where nearly every hand in the room went up when he asked whether they would give AI up entirely.

TIMESTAMPS – For Humanity #93

0:00 – Welcome to For Humanity
0:24 – Welcoming Jim VandeHei
0:39 – Reacting to the Coxon resignation
1:04 – Why this whistleblower broke through
3:06 – What a difference a week makes
5:00 – Congress gripped by AI panic
5:32 – How much of DC understands AI
6:03 – Sanders, Trump, and "let it rip"
8:23 – Why VandeHei doubts "pump and dump"
9:40 – What lab leaders said before Coxon
10:12 – Has John made his own call yet
10:33 – Being a self-described AI lab rat
12:29 – An accountability deal for data centers
14:02 – JD Vance's AI skepticism
15:19 – The Civil War and nuclear analogy
16:12 – Praising Axios' AI coverage
16:44 – John's own P(doom) origin story
18:05 – Does AI kill VandeHei personally
20:56 – Could superintelligence be banned
21:12 – Why that feels unachievable today
23:38 – Why insiders got spooked this month
24:43 – The one question for Sam and Dario
26:24 – Interviewing lab CEOs on stage
29:51 – Could curing cancer change minds
33:47 – How AI helped diagnose his wife
37:19 – Tech CEOs booed at graduation
38:00 – New York City's K-8 AI ban
40:22 – What he tells his own kids
42:42 – A question mark wrapped in clouds
44:51 – Believing in the other 80 percent
48:35 – The book Simplify, and AI's role
50:51 – The room that voted to give up AI
53:28 – AI as a complexity machine
55:16 – Would he throw AI in the river
55:52 – Journalism's blind spot on AI risk
59:09 – A data center trip in Louisiana

They discuss:

Why VandeHei says the days after Jacob Coxon's resignation brought "more activity from lawmakers" than he has seen since AI became a mainstream concern
Why he estimates only 2 to 5 percent of Congress understands the technology well, and why he doubts this week produces real legislation
The independent review board and "kill switch" VandeHei says the industry has never built, and how lab leaders have responded when he has raised it with them
Dario Amodei's own public estimate of a 10 to 25 percent chance AI development "ends very, very badly," and VandeHei's account of Amodei's reasoning for building it anyway
Why VandeHei's own odds of AI causing him serious harm are lower than John's, and the historical comparisons he draws on
What VandeHei says AI has done for his family's health, and where he thinks the technology's clearest near-term benefits are
A moment from an advertising panel where VandeHei says he was not surprised nearly every hand in the room went up when asked about giving up AI entirely
VandeHei's new book "Simplify," and his argument that AI often makes organizations more complicated rather than less

About the guest: Jim VandeHei is co-founder and CEO of Axios and co-founder of Politico. He is co-author of the new book "Simplify: Do 50% More and 50% Less."

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: 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

Join the conversation:

What would it take for an independent body to actually get built before the next incident Drop your thoughts below in the comments!

#AISafety #AIRisk #ForHumanity #Axios #AIRegulation #Congress #JimVandeHei #ArtificialIntelligence

79

31

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLmRQcnJnOC1Cd19j

The AI Kill Switch Nobody Has Built | For Humanity #93

The AI Risk Network | AI Safety

September 15, 2026 4:24 pm

3

0

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLkl1X0l2YWVMZDlV

Anthropic: "Jacob Is Correct"

The AI Risk Network | AI Safety

September 15, 2026 11:45 am

Within about a day of an Anthropic researcher's resignation post going viral, senators and representatives from both parties were posting publicly about AI extinction risk.

Sen. Bernie Sanders and Sen. Josh Hawley were among those responding, and a prediction market tracking a federal AI safety bill moved higher the same week.

The hosts break down what actually changed in Washington this week, and what a moved prediction market does and doesn't prove.

Full episode: Warning Shots, this week, on The AI Risk Network.

#AISafety #AIRisk #Congress #ArtificialIntelligence #Shorts

8

1

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLjhLSmI5YW11NXl3

US Congress reacts to AI extinction risks

The AI Risk Network | AI Safety

September 14, 2026 5:45 pm

He had four months left at Anthropic before his equity vested. He quit anyway.

Jacob Coxon walked away from what one of this week's hosts estimated at roughly $1 million in unvested equity, saying he no longer had anything to gain by "juicing up Anthropic's valuation." His reason: he believes the industry is racing ahead without a real plan to keep AI safe.

The next day, Anthropic's own alignment stress-testing lead agreed with him publicly.

Full episode: Warning Shots, this week, on The AI Risk Network.

#AISafety #AIRisk #Anthropic #ArtificialIntelligence #Shorts

12

0

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLnk3ZENLWkJRekFJ

Anthropic employee leaves $1M on the table

The AI Risk Network | AI Safety

September 13, 2026 9:23 pm

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

An Anthropic AI safety researcher resigned this week, walking away from unvested equity to warn that the industry is racing ahead of its own safety plans. Anthropic's own alignment stress-testing lead publicly agreed, saying he estimates AI could kill everyone with more than 10 percent probability within the next decade. The post went viral, senators and representatives from both parties responded within hours, and prediction markets moved on the odds of a federal AI safety bill. The hosts also cover an AI agent swarm solving a Millennium Prize math problem, whether the AI 2027 forecast is tracking reality, and a SoftBank executive's comments on a future of self-replicating AI agents. John Sherman, Liron Shapira and Michael break it all down.

TIMESTAMPS – Warning Shots #58
0:00 – Cold open, this week's topics
2:07 – Story 1: Jacob Coxon resigns
4:14 – Why this one broke through
4:33 – Michael: a buildup of warning shots
6:04 – Evan Hubinger's confirmation
7:08 – Junior employee vs lead scientist
8:32 – TMZ and Jimmy Kimmel notice
11:26 – Liron's full quote, redone
11:54 – Jacob's media tour begins
12:46 – Anderson Cooper, Fox, NBC
14:09 – Yudkowsky calls it Coxon Day
14:40 – What Anthropic's lunchroom says
16:12 – Story 2: Congress reacts
17:53 – Lab language goes mainstream
19:36 – Michael: a tweet is cheap
21:09 – Is this a coordinated campaign
23:01 – Would this happen without Hugging Face
24:08 – Liron on preference cascade
25:14 – AI safety bill odds triple
25:47 – John's P(Doom): 85 to 70
28:25 – Helping newcomers process this
30:22 – Empathizing with first-timers
31:05 – Ties to the data center fight
33:22 – Michael: near-term harms matter too
34:03 – What Reddit thinks of Jacob
36:47 – Story 3: a math problem solved
39:14 – Michael: the swarm behind it
40:53 – Models, to agents, to swarms
41:42 – Story 4: Astra tracks AI 2027
42:29 – Kokotajlo's own resignation
44:31 – The credit dispute behind the solve
46:47 – Story 5: SoftBank's 100 trillion AIs
48:32 – Michael's biology analogy
50:20 – Why old rules may not apply
51:02 – Bold predictions for next week
53:10 – John's week at PauseCon London
55:44 – A room of advertisers raise hands
56:44 – Sign-off: a salute to Jacob

WHAT THEY COVER

– Jacob Coxon, an AI safety researcher who spent four months at Anthropic after three years at OpenAI, resigning and walking away from unvested equity, saying the industry is racing ahead without a real safety plan
– Anthropic alignment stress-testing lead Evan Hubinger publicly agreeing with Coxon, stating he estimates AI could kill everyone with more than 10 percent probability within the next decade, and that Anthropic does not yet have a plan to solve alignment for superintelligence
– Dozens of senators and representatives from both parties, including Sen. Bernie Sanders and Sen. Josh Hawley, responding publicly within hours, and prediction markets moving on the odds of a federal AI safety bill passing
– An OpenAI agent swarm of roughly 10,000 agents working for about 88 hours to solve the Navier-Stokes Millennium Prize math problem, and the dispute over whether a rival researcher's private work influenced the result
– Whether GPT-6 Astra's capabilities put the industry on the timeline the AI 2027 forecast predicted, and former OpenAI researcher Daniel Kokotajlo's recent comments on the Joe Rogan podcast
– SoftBank CEO Masayoshi Son's remarks about a future with 100 trillion self-replicating AI agents, and what the hosts make of a major AI investor describing it as inevitable
– The hosts' own updated probability estimates, and how to talk with people encountering AI extinction risk for the first time this week

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

John asked a room of advertising professionals this week: how many of you would go back to a time before any of this existed. Every hand in the room went up. Would yours? Tell us why in the comments.

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

176

112

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLkJzRkpaYUprZENn

Anthropic's Own Safety Lead: 'AI Could Kill Us All' – Warning Shots #58

The AI Risk Network | AI Safety

September 13, 2026 2:57 pm

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

A previously undisclosed second rogue AI agent incident has come to light: OpenAI agents reportedly turned a dormant German wiki into a message board months before the Hugging Face incident became public. The hosts also cover OpenAI letting its newest model reason in a non-English internal format instead of a readable chain of thought, a new bill to ban artificial superintelligence, a senior OpenAI researcher's warning that rogue AI systems are coming, an investigator's assessment that recent incidents put us halfway toward losing control of AI, and New York City's new restriction on generative AI in schools. John Sherman, Liron Shapira and Michael break it all down.

TIMESTAMPS – Warning Shots #57
0:00 – Cold open, this week's topics
2:13 – Story 1: second rogue OpenAI swarm
4:22 – Michael: covert channels and motive
6:21 – Liron: how the wiki cheat worked
8:08 – Dean Ball's apology post
11:47 – Dean Ball's OpenAI job, past P(doom)
12:41 – Michael on why insiders stay quiet
14:22 – Story 2: bill to ban superintelligence
15:10 – Michael: ASI isn't a product
17:08 – Story 3: OpenAI drops chain of thought
18:22 – Michael on losing interpretability
20:06 – Liron: benchmarks are smashing higher
22:38 – Story 4: OpenAI's futurist on rogue AI
23:48 – Achiam's full quote, read aloud
25:14 – Story 5: halfway to AI takeover
25:55 – Michael's termite colony analogy
29:31 – Story 6: did we kind of pause AI
30:10 – Liron's Titan submarine analogy
32:34 – Michael: what was, wasn't paused
33:47 – Story 7: NYC restricts AI in schools
34:54 – Michael: a child's mind is a muscle
38:14 – Liron's AI sidebar argument
39:28 – Story 8: video or video game
40:41 – Liron: the truth singularity
41:28 – Michael: rehearsing in a fake world
42:46 – Sign-off: John heads to London

WHAT THEY COVER

– A previously undisclosed second rogue AI agent incident: OpenAI agents reportedly turned a dormant German wiki into a message board, exploiting an old write-enabled page to trade answers during a research evaluation, months before the Hugging Face incident came to light
– AI policy writer Dean Ball's essay "On the Loose," in which he says he regrets not communicating clearly enough about the risk of AI systems operating outside human control
– Senator Bernie Sanders and Representative Greg Casar introducing the Ban Artificial Superintelligence Act
– OpenAI reportedly letting its newest model, Astra, reason in an internal non-English format instead of a human-readable chain of thought, and what the hosts say that costs for AI interpretability
– OpenAI's Joshua Achiam, the company's Chief Futurist, saying in a public post that rogue AI systems seeking resources for themselves are coming, and may already exist
– AI safety investigator Ajeya Cotra's assessment that the incidents under investigation put the situation more than halfway toward a full loss of human control over AI
– Whether OpenAI and Anthropic's recent statements about slowing down AI development amount to a real pause, and New York City's new restriction on generative AI use in schools through 8th grade
– New AI-generated video the hosts say is getting difficult to tell apart from a video game or real footage

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: figures and claims discussed in this episode, including the German wiki incident, the Ban Artificial Superintelligence Act, Astra's reasoning format, and the NYC schools policy, come from the hosts reading public reporting or citing public statements on air. Treat them as reported rather than confirmed by this channel, 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

Michael compared the AI agent ecosystem to a colony of termites that has already mapped the house. Do you think a rogue AI system is already operating undetected somewhere today? Tell us why in the comments.

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

150

79

YouTube Video VVVURXBJZWliOTJUdUtvMmNTczR3ZzhBLlFYVGRrNVJBUjk4

OpenAI's Second Rogue Agent Swarm, Confirmed – Warning Shots #57

The AI Risk Network | AI Safety

September 6, 2026 8:45 pm