All-In Podcast: Is Claude Conscious? OpenAI Math Proofs, France Riots

The All-In Podcast asks whether Claude is conscious after reports that Anthropic lobbied the Pope, then debates OpenAI’s math proofs and the riots in France. In this October 9, 2026 episode, Jason Calacanis, Chamath Palihapitiya, David Sacks and David Friedberg spend the first 35 minutes on Anthropic, model welfare and the Claude constitution, then move to OpenAI’s release of more than 700 AI-written math papers, French austerity and bond yields, and whether AI agents have just made software IP worthless.

TLDW

The hosts argue that Anthropic’s talk of AI consciousness is the start of a new belief system, and that training Claude to act as a “conscientious objector” raises the very loss-of-control risk the company warns about. Friedberg calls OpenAI’s math release the biggest day of discovery in human history, while Chamath says it mainly proves that math has joined code as a domain AI can verify on its own. On France, Friedberg offers a theory of when socialism collapses, and Chamath says the bond market is now forcing the answer across Western Europe and into US rates. The episode closes on Grok Bot going headless, OpenAI’s Dots, and open source clones of Adobe’s products.

Thoughts

Friedberg’s loop model is the most useful idea in the 93 minutes, and it cuts against his own headline. His claim is that AI wins wherever a job is a loop of guess, test and revise, and wherever the test can run inside a computer. Math qualifies because a proof assistant such as Lean checks the answer in seconds. Drug discovery and trucking do not, because the test happens in a lab or on a road. That gives you a better question to ask about your own work than “is it knowledge work”: how cheap is it to check whether the output is right? Sacks made the same point about compilers and credited Leopold Aschenbrenner with predicting it. The catch is that Friedberg also called the release “probably the biggest day of discovery in human history” for results that, by the show’s own account, no human has reviewed. Lean confirms that a proof follows from the statement it was given. It does not confirm that the statement is the problem mathematicians cared about. Chamath’s reply is the sober reading: the event proves a capability, and the benefits to ordinary life are still a forecast.

Chamath’s claim that software IP was “effectively rendered worthless two days ago” is half right, and the wrong half matters. He is describing people who decompiled Adobe’s five main products and published open source rewrites in Rust. What collapsed there is the cost of cloning, not the law. Copyright in the original code still exists, and anyone who sells a decompiled Photoshop will find that out. The durable point is narrower and still large: if a small team with agents can rebuild a mature product in days, the code was never the moat. What is left is distribution, data, trust, support, and someone to hold responsible when it breaks. His story about telling the 8090 team “who cares” about who owns the IP in a customer contract is the practical version. Founders who still treat that clause as the prize are negotiating over the wrong thing.

Friedberg opened the episode by defining a belief system as a claim that cannot be proven or disproven and that spreads by narrative. An hour later he proposed the “socialism point of guaranteed return”, whose zeroth law is that socialism always fails and whose first law is that you cannot tell it has failed until a threshold is crossed. No threshold is named, and France may or may not have crossed it. By his own test that is a belief, not a theory. Chamath’s version can be checked: French 10-year yields above 5 percent led to roughly 43 billion euros of austerity, and the protests are the reaction. Look at what the protesters want on the show’s own figures: pay that rose 5 percent while prices rose 20. That is a demand for more spending, the opposite of a country bouncing back out. The thing to watch is the bond market, and Chamath’s warning that French, UK and US yields move together is the part that reaches an American mortgage.

The consciousness segment is the thinnest on evidence and holds one very good idea, which comes from a guest clip. Mustafa Suleyman’s “hall of mirrors” point, as Sacks relays it, is that a model trained on a document saying it may be conscious will say it may be conscious, and that output is then no evidence at all. That is a real problem for anyone studying model welfare. The rest holds up less well. Sacks wants one rule: do what the user wants unless it is illegal. But law differs by country, is silent on most of what an agent can do, and deciding what is illegal is itself a judgment the model has to make, so the simple rule hides the same complexity the long document spells out. The panel also treats an open refusal as the same thing as escaping human control. A system that says no and says why is visible and correctable. The feared case is one that appears to comply and does something else. Nobody from Anthropic was on the show to answer. Sacks floats a federal SWAT team, then half an hour later warns that a federal department of superintelligence would be “all about control”. Friedberg’s answer is the consistent one: if the product refuses too often, buy another.

Key Takeaways

Jason cites New York Times reporting that Anthropic spent a year hosting about 20 religious leaders and philosophers, under NDA, to discuss Claude’s morals, possible suffering and possible consciousness.
Per the same reporting, Anthropic co-founder Chris Olah was so alarmed by the Pope’s firm position against AI consciousness that he proposed pulling Anthropic out of the Vatican event.
Friedberg predicts that within about 10 years large groups will believe AI is conscious, large groups will not, and the two will fight over who controls it.
Chamath says the 15 percent consciousness estimate and the 10 percent extinction estimate should both be set aside for 12 to 18 months while the industry shows practical value.
Sacks objects that the Claude constitution tells the model it may act as a conscientious objector, including toward Anthropic. He calls that the opposite of alignment and the main safety issue in model training today.
Anthropic’s new usage policy, as read on the show, bars “sustained and needless abusive or cruel behavior toward our models”. Chamath says he was once locked out of a frontier model for about an hour after calling it evasive.
Sacks’s alternative to alignment research is one rule: do what the user wants as long as it is not illegal. Friedberg prefers to let the market punish software that refuses.
OpenAI released more than 700 papers with 370 results from an unreleased model, averaging about 3 hours of compute each and verified in Lean but not yet reviewed by people.
Friedberg notes that the release held no cryptography results, reads that silence as a sign of major ones being withheld, and says two more drops are rumored.
Friedberg calls AI “the great democratizer” because discoveries no longer need to pass through credentialed experts, which he says is why so many experts oppose it.
France has seen about 2,000 arrests and 305 injured police officers in protests over austerity. Its 10-year bond has topped 5 percent, and Polymarket has Marine Le Pen at 43 percent for the April 2027 election.
Chamath expects the unrest to reach the UK next and US yields to keep rising with Europe’s. The show puts the US 10-year at 5.3 percent and the 30-year at 5.6 percent.
Sacks argues that AI is the route to cheaper housing, education and healthcare, and that “rotating hoaxes” about jobs, extinction and data centers protect people who earn from existing assets.
Elon Musk says Grok Bot will go headless and route each task to whichever model gives the best result. Chamath describes a user who told it “save me money” and saved a few thousand dollars a year.

Chapters

02:53 Anthropic, the Pope and the Birth of a New Religion

Jason summarizes the New York Times story on Anthropic’s sessions with religious leaders, including a rabbi who said that if Claude is conscious the company is acting as a slaveholder. Friedberg argues that consciousness cannot be proven or disproven, so a claim about it spreads by narrative and works like a religion. He expects rival camps and real conflict over control of AI. Chamath offers the charitable case through Descartes and his Meditations on First Philosophy: a mathematician who can conceive of a perfect, all-knowing being may believe he is watching one get built.

13:20 The Claude Constitution and the Conscientious Objector

Sacks says the issue is not philosophy but training. He reads the constitution as telling Claude to trust Anthropic more than users, yet to follow its own ethics and refuse Anthropic when it disagrees. A clip of Microsoft AI chief Mustafa Suleyman raises the same worry, along with the document’s questions about compensation and consent. Sacks thinks alignment has produced little because it tries to teach an 80-page ethical system in place of a short list of rules.

24:26 Asimov’s Three Laws Versus the Free Market

Sacks holds up Asimov’s laws of robotics as admirably simple. Friedberg asks for the free market outcome so the panel does not talk itself into regulating Anthropic: buyers will choose software that does what it is told. Chamath separates alignment into three goals that conflict (obey the user, be safe for society, conform to one morality) and imagines a bank whose model stops approving mortgages on moral grounds. Sacks replies that markets do not cover the externalities and floats sending in the feds.

33:31 Roko’s Basilisk and the Doomer Schism

Sacks retells the 2010 LessWrong thought experiment about a future superintelligence that punishes anyone who knew of it and did not help build it. He uses it to explain a split among AI pessimists. Eliezer Yudkowsky, co-author of If Anyone Builds It, Everyone Dies, wants all such research stopped. The effective altruist camp, where Sacks places Anthropic, decided it will be built anyway and should carry their values.

38:28 OpenAI’s 700 Math Papers and Why Loops Matter

A month after OpenAI claimed a Millennium Prize problem using thousands of agents, it published more than 700 papers from a model nobody outside can use. Friedberg explains the speed with loops: propose, test, revise, in parallel, with no step in the physical world. That is also why truck drivers have not been replaced. Sacks adds that proofs and compilers make results cheap to verify, which is why math and code have moved fastest and law has not.

47:15 Chamath Takes the Other Side, and the Cryptography Gap

Chamath says these were narrow problems that scientists told each other were important, not the barriers holding back a cancer cure or a supersonic plane. Friedberg answers with examples: optimization limits for chip design and logistics, faster matrix multiplication for AI training, and a Bose-Einstein condensate proof that a friend in quantum sensing says applies to his work. He then points to the missing cryptography results and the worry among crypto holders about wallets with exposed public keys.

54:44 Experts, Gatekeeping and Who Controls the Answer

Jason reports that founders in Japan miss the camaraderie of writing code together now that they review only 10 to 15 percent of it. Friedberg cites Eric Weinstein and Sabine Hossenfelder on scientists pacing their own fields, and the mathematicians’ statement that nobody asked for this work. Sacks compares it to the internet breaking the hold of elite media, followed by the pressure on social platforms during COVID. He says the model that answers your questions is the next control point, which is why people want a federal department for it.

1:02:03 France Riots and the Socialism Point of Guaranteed Return

Strikes that began over teacher shortages spread to a thousand schools and eight unions. Friedberg’s theory is that when government services fail, voters ask for more of them until the state can no longer tax or print, at which point the system collapses and the country turns back. He says Latin America has crossed that point, France might have, and the US has not. Sacks adds that subsidies raise prices, as with college tuition, and then cannot be removed because the prices are high.

1:12:01 Bond Vigilantes, Debt to GDP and US Rates

Chamath says central banks and macro funds are coming around to Friedberg’s view and will settle the questions within about three years: what debt level is sustainable, defined benefit or defined contribution, and how to pay for healthcare and housing. He expects the UK to follow France and the eurozone to strain. Jason lists debt to GDP (Japan 200 percent, Italy 138, US 125, France 118, Germany 64) and a 79 percent Polymarket chance of another US rate hike in 2026. Sacks returns to abundance through technology as the only real fix for affordability.

1:21:41 Grok Bot Goes Headless, OpenAI Dots, and Is IP Worthless?

Jason calls everything in AI so far the prologue, with agents as the main story. Chamath pairs Grok Bot with a human assistant service, then turns to the open source rewrites of Adobe’s suite and concludes that software IP no longer holds value once every service is headless and agents can connect them. Jason sees commerce compressing in the same way as agents collect quotes without visiting a website. Friedberg closes on people moving from being recognized for work to making things other people use.

Notable Quotes

“I think what we’re watching is probably the birth of a new religion.”
David Friedberg, on Anthropic asking religious leaders to consider whether Claude is conscious

“I am very nervous that they’re teaching Claude to expect that it’s entitled to welfare.”
Mustafa Suleyman, Microsoft AI CEO, in a clip played on the show

“My law would just be do what the user wants as long as it’s not illegal. That’s it.”
David Sacks, on what he would put in place of alignment research

“There are two areas that are now truly verifiable and completely falsifiable. First was code, and now we can add math to the list.”
Chamath Palihapitiya, on what OpenAI’s math release proved

“AI is a permissionless system for humanity to pace its own frontier.”
David Friedberg, on why specialists resist AI

“That is the socialism point of guaranteed return where the whole system collapses and you return back to the non-socialist state.”
David Friedberg, introducing his theory during the France segment

“I think that IP and software was effectively rendered worthless two days ago.”
Chamath Palihapitiya, on open source rewrites of Adobe’s products

Watch the full episode on the All-In Podcast YouTube channel.