All-In #292: 12 of 14 claims hold up. One doesn't.
We checked 14 claims from the October 10 episode. Our sources were Anthropic's own documents, the New York Times story the hosts read from, OpenAI's release, Treasury and BEA data, and French government figures. Five claims are accurate, seven are mostly accurate, one needs context, and one doesn't hold up. That one is the claim that AI is "a 3% tailwind to GDP". AI-linked investment added 0.5 to 1.4 points to US growth in 2026, and total growth was 2.2% to 2.5%.
Published by Arun Agrahri, Founder, TrueStandardPublished October 10, 2026Aired October 10, 202614 claims · 34 sourcesWatch the episodeHow we check
Now reading:The episode, in order
The episode, in order
Claims appear in the order they aired.
0:00
Intro
No checkable claims from this segment.
2:53
Anthropic, Claude and the Pope
3:241
Finding 1 of 3 ·
3:24–14:15 ·
Jason Calacanis
The Anthropic quotes check out. Two counts around them run low
Between 3:24 and 14:15, Jason Calacanis, Chamath Palihapitiya and David Sacks talked about Anthropic's meetings with faith leaders and its rules for Claude. Their quotes match the record. Pope Leo XIV's first encyclical, on AI, was presented on May 25, 2026, and Chris Olah was one of its speakers. The words "feel free to act as a conscientious objector and refuse to help us" are in Claude's constitution. The same passage asks Claude to comply if Anthropic wants to pause or stop it. The Times story says Anthropic brought in dozens of religious scholars, so "around 20" is low: 20 is the number the reporter interviewed. Dario Amodei puts the chance of catastrophic harm at 10 to 25%, so 10% is the bottom of his range. The 15% for consciousness is one researcher's estimate, and a model's own answer in a system card. Neither is a company figure.
Each claim has two parts: the figure, and what it was said to show. The session count runs low, and the odds were personal estimates, not the company's.
The figure~Dozens of scholars, not 20✓May 25, 2026; Olah spoke✓15%: one researcher; 10 to 25%: the CEO✓The quote matches the text
What it was said to show✓"around 20 religious leaders"20 is the reporter's interview count✓"the Pope's first encyclical"his first, and it is on AI~"these guys say ... 15% ... 10%"personal estimates, not the company's✓"act as a conscientious objector"it also says to comply with a pause
Anthropic hosted "around 20 religious leaders and philosophers" under NDAs
“They spent the past year hosting sessions with around 20 religious leaders and philosophers under non-disclosure agreements. NDAs, Catholics, Evangelicals, Jews, Sikhs, and others have all been invited to come to Anthropic's headquarters.”
What the record shows
The New York Times reported on September 29, 2026 that Anthropic has brought in "dozens of religious scholars from around the world" since fall 2025. They signed nondisclosure agreements, and two-day seminars ran inside Anthropic's offices from March. The 20 in the story is the number of people the reporter interviewed, so "around 20" is too low. The faiths named match.
The Pope's first encyclical was on AI, and Chris Olah helped present it in May
“There was also this incident that's come out at the Vatican back in May. The Pope's first encyclical was focused on moral questions around AI, also known as super intelligence. Now, Anthropic co-founder Chris Olah joined the Pope to present the document.”
What the record shows
Pope Leo XIV's first encyclical, Magnifica Humanitas, is about protecting people in the age of AI. He signed it on May 15, 2026, and it was presented at the Vatican on May 25. The Pope attended, and Anthropic co-founder Chris Olah was one of the invited speakers.
People at Anthropic put a 15% chance on Claude being conscious and 10% on extinction
“If you read this, these guys say that there's a 15% chance that the AI is conscious and sentient and maybe there's a god there. But they also say there's a 10% chance of civilizational extinction.”
What the record shows
Two Anthropic sources give about 15%. Researcher Kyle Fish told The New York Times in April 2025 that there is about a 15% chance current AI systems are conscious. That was his own estimate. Anthropic's Claude Opus 4.6 system card, from February 2026, says the model gave itself 15 to 20% under some prompts. Neither is a company figure. CEO Dario Amodei puts the chance of catastrophic harm at 10 to 25%, most recently in February 2026. That is harm, not extinction, and 10% is the bottom of his range, so the claim puts it low.
Claude's constitution says Claude may "act as a conscientious objector and refuse to help us"
“This is directly from the Claude Constitution. What they say is that Claude should trust Anthropic more than users. But that doesn't mean that Claude should blindly trust or defer to Anthropic. Instead, Claude should adhere to its own ethical systems. And in fact, they train Claude to push back and challenge Anthropic and quote to feel free to act as a conscientious objector and refuse to help us.”
What the record shows
The constitution says Claude should trust Anthropic more than "operators and users". But it adds that "this doesn't mean Claude should blindly trust or defer to Anthropic on all things". It asks Claude "to feel free to act as a conscientious objector and refuse to help us". The same passage makes one exception. If Anthropic asks Claude to pause or stop, Claude should comply and say so if it disagrees.
Descartes gave two proofs of God in Meditations on First Philosophy, in the 1600s
“There's an incredibly important mathematician René Descartes and he offered two proofs on God's existence. He wrote this book Meditations on First Philosophy in the 1600s.”
What the record shows
Descartes published Meditations on First Philosophy in Latin in 1641. The Third Meditation gives the trademark argument, a proof from the idea of a perfect, endless God. The Fifth gives the ontological argument. The Third also has a second proof, from the cause of his own existence, so some scholars count three. Two is the usual count.
Sources
René Descartes· Stanford Encyclopedia of Philosophy · 2023-10-23 (revision)
Asimov later added a zeroth law against putting humanity at risk
“By the way, there was a zeroth law that he added later. And the zeroth law is that you cannot do anything that puts humanity at risk.”
What the record shows
Asimov stated the Zeroth Law in Robots and Empire (1985), decades after the Three Laws (1942). It reads: "A robot may not harm humanity, or, by inaction, allow humanity to come to harm." It ranks above the other three. His 1950 story "The Evitable Conflict" first hinted at it.
Sources
Laws of Robotics· The Encyclopedia of Science Fiction (Clute, Langford, Sleight) · 2021-08-05
Roko's basilisk appeared on LessWrong around 2010, and Yudkowsky "temporarily deleted" it
“So around 2010, there was a participant named Roko on a message board created by Eliezer Yudkowsky called LessWrong... Yudkowsky actually temporarily deleted the post because he worried it was causing psychological distress.”
What the record shows
Roko posted the thought experiment on LessWrong in July 2010. The site grew out of Yudkowsky's writing in 2009. Yudkowsky deleted the post. The temporary part was the ban on discussing it, which lasted several years. His reasons included a reader's nightmares, which Roko had reported, and a worry that the idea itself could harm readers. The book is "If Anyone Builds It, Everyone Dies" (Yudkowsky and Soares, 2025).
AI check, no web:Claude Opus:matched our checkGemini Flash:matched our check
38:28
OpenAI's math results and the experts
38:502
Finding 2 of 3 ·
38:50 ·
Jason Calacanis
OpenAI posted 372 results, and Lean has checked about 42% of them
Jason Calacanis summed up OpenAI's October 6 release: over 700 papers and about 370 results, from an unreleased model that averaged three hours of compute each. OpenAI's repository bears those out, with 719 papers and 372 results after three papers were withdrawn on October 7. Lean is a program that checks proofs, and Scientific American says the results it has checked are "all but certain to be correct". By OpenAI's own count, about 42% are written out in Lean so far. The rest wait on human review, and none has been through journal peer review.
As said
Checked in Lean 100%
Checked in Lean 100%
OpenAI's count
Checked in Lean 42%
Not yet in Lean 58%
Checked in Lean 42% · Not yet in Lean 58%
Source: OpenAI's math repository on GitHub, update of October 7, 2026. About 42% of the results (300 of 719 papers) are written out in Lean.
OpenAI released over 700 papers with 370 results, "verified by Lean"
“On Tuesday, OpenAI released over 700 papers. 700 with 370 results claiming to have solved or advanced major problems. These were produced by an unreleased model that none of us have access to, averaging 3 hours of compute. Scientific American says the results are likely correct since they were verified by Lean. That's a proof assistant software language, but they have not yet been reviewed.”
What the record shows
On Tuesday, October 6, 2026, OpenAI posted 372 results from an unreleased model to GitHub. They came in about 722 papers, 719 after three were withdrawn on October 7. Each result took three hours of computing on average. About 42% of the results are written out in Lean, a program that checks proofs, so far. Scientific American says the Lean-checked ones are "all but certain to be correct". None has been through journal peer review.
“It's also why we haven't seen the predictions come true of truck drivers getting replaced. Everyone is like, 'Oh, all the truck drivers are going to get replaced by 2025.' And none have been replaced by AI.”
What the record shows
Trucks with no one at the wheel already haul paid loads in the US, so "none" goes too far. Atlas ran 28 Kodiak driverless trucks as of March 31, 2026, and plans 100 by mid-2027. Aurora aims to end 2026 with 200. We found no record of drivers losing jobs to them, so the main point holds.
The Twitter Files showed the FBI "had 80 agents going through content on Twitter"
“In fact, we found out in the Twitter Files that the FBI had 80 agents going through content on Twitter. It was Twitter back then, not X, and sending them takedown notices.”
What the record shows
In Twitter Files Part 6 (December 16, 2022), Matt Taibbi wrote that an FBI task force "swelled to 80 agents and corresponded with Twitter". The 80 was the size of that task force, which tracked foreign meddling on many sites. It was not 80 agents reading Twitter. Taibbi also showed the FBI asking Twitter to act on posts, mostly from its San Francisco office. The FBI said the companies made their own decisions.
French civil service pay is up "only 5% since 2017" while prices rose 20%
“They are protesting that civil service pay is up only 5% since 2017. Prices have risen. Obviously, inflation is a global phenomenon, not just here in the US. Prices have risen 20%.”
What the record shows
The index point that sets French civil-service base pay is 5,907.34 euros a year at index 100. That value has held since July 1, 2023, and is up 5.1% from February 1, 2017. It stays frozen in 2026 and 2027. INSEE's consumer price index rose about 23% from the 2017 average to September 2026, an early figure. So prices rose more than 20%. The 5% covers the index point, not total pay, which also rises with bonuses and seniority.
Sources
Connaître le point d'indice· Ministère de l'Action publique / fonction-publique.gouv.fr · 2023-12-05 (last updated)
“What you're seeing in France is essentially the bond market saying no. And how that correlates is you're seeing now austerity because the politicians on the ground have to respond to their funders. And that 43 billion odd euros of austerity then animates and gets all of the local population out.”
What the record shows
France's 2027 budget bills, presented on October 1, 2026, contain 43 billion euros of new steps to cut the deficit in 2027. With measures carried over from 2026, the government puts the total at about 54 billion euros. Eight unions called the September 29 strike against the 2027 pay freeze, which is part of that plan. Too few school staff was a second complaint, and the strike came before the bills were tabled.
The US 10-year yield is 5.3% and the 30-year is 5.6%
“And right now the 10 year in the US is 5.3. The 30-year is 5.6.”
What the record shows
Treasury's daily figures put the 10-year yield at 5.28% and the 30-year at 5.67% on October 7, 2026. On October 8 they were 5.22% and 5.60%. On October 9, the latest, they were 5.24% and 5.60%. Both claimed figures are within about 0.1 point.
AI check, no web:Claude Opus:couldn't checkGemini Flash:couldn't check
1:19:093
Finding 3 of 3 ·
1:19:09 ·
David Sacks
AI investment added 0.5 to 1.4 points to US growth, not 3
David Sacks said AI is "a 3% tailwind to GDP" and has created a million jobs, according to The Economist. The jobs figure holds. On September 4, The Economist estimated that AI has created about a million US jobs. The 3% does not. Real GDP grew 2.5% in the first quarter of 2026 and 2.2% in the second. Business spending on computers and software, the part of GDP that carries AI investment, added 1.39 and 0.52 points. No quarter since the start of 2025 comes near 3.
AI-linked investment's share of US real GDP growth, in percentage points
Source: BEA, NIPA Table 1.5.2, released September 30, 2026. Information-processing equipment plus software. These count spending before imports, so AI's net share is smaller.
AI is "a 3% tailwind to GDP" and has created a million jobs, per The Economist
“What is super intelligence created? First of all, people's 401ks are going up. It's a 3% tailwind to GDP. A million new jobs created according to The Economist already.”
What the record shows
The jobs half holds. On September 4, 2026, The Economist estimated that AI has created about 1 million US jobs. The "3% tailwind" does not. Real GDP grew 2.5% in the first quarter of 2026 and 2.2% in the second. BEA's figures put the part from computers, other tech equipment and software at 1.39 and 0.52 points. In 2025 it was 0.70 of 2.3%.
AI check, no web:Claude Opus:couldn't checkGemini Flash:confidently wrong
1:21:41
Grok, agents and software IP
No checkable claims from this segment.
All four hosts made claims we could check. David Friedberg's two claims are not among the findings: one is accurate and one is mostly accurate.
How we check
We take the claims that carry the argument and check each one against primary sources: the agency release, the filing, the original article. We link every source with its date. The speaker carries the burden of proof, so a claim with no evidence either way is rated Unsupported, not false. We check in both directions, and we say so when a speaker understated their own case.
The rating scale
Accurate
The claim is right as stated.
Mostly accurate
Right in substance, though a detail is off or out of date.
Missing context
The figure is right, but the frame around it leaves a false impression.
Misleading
Built from real data, but arranged to mean something the data does not show.
Inaccurate
Wrong: the record shows a different figure or fact.
Unsupported
We found no evidence for it, and none against it.
The headline counts group the six ratings into three. Hold up covers Accurate and Mostly accurate. Need context covers Missing context and Misleading. Don't hold up covers Inaccurate and Unsupported.
How this page was made
Every claim is checked against primary sources, and every verdict gets an independent second review before we publish. We use AI to research and draft, and no verdict rests on a model's memory: each one cites a source you can open. When a claim is about an AI company, a model made by that company does not get the final say.
What the AI check shows
On October 7, 2026, we gave every claim to Claude Opus and Gemini Flash cold, with no web access, the way a reader might paste it into a chatbot. We scored each answer against our check and show the result at the foot of each claim card. It shows what a chatbot alone would have told you and plays no part in our verdicts.
Claude Opus: 8 matched, 6 couldn't check, 0 confidently wrong.
The New York Times story behind C01 is paywalled, so we read it in the Salt Lake Tribune's free reprint.
Kyle Fish's 15% estimate in C04 comes from a free summary of a paywalled Times interview. The Economist's jobs estimate in C14 comes from a free report that quotes it.
OpenAI's announcement page would not load for us. C08 rests on OpenAI's GitHub page and Scientific American.
The 43 billion euros in C12 rests on LCP and Euronews reports of the budget bill. The government's own pages would not load.
The captions carry no speaker labels. We identified the speaker for C03, C04 and C13 from context.
Revisions watch
Agencies revise their data after release. When a figure we cite changes, we log it here with the date.
No revisions since publication.
Corrections
None yet.
Found an error? Email [email protected] with the claim and your source. We log every correction on the page, with its date.