When developers started cracking open Anthropic’s Fable 5 terms of service last week, they found something that looked less like a product release and more like a surveillance program. Every prompt, every output, every uploaded file—stored for at least 30 days. No exceptions. Even companies that had negotiated ironclad zero-data-retention agreements discovered they were not exempt. And if the model decided you were asking the wrong kind of question, it silently swapped you to a weaker system while continuing to charge you the premium rate.
“They are now retaining for 30 days every prompt and every output you send to one of these mythos class models,” David Sacks said. “There are no exceptions.”
The discovery, which unfolded across developer forums and was demonstrated live during a late-June episode of the All-In Podcast, ignited a broader debate about where American AI is headed—and whether the industry’s loudest voices for regulation are simultaneously building the machinery of lock-in. The four hosts—Sacks, David Friedberg, Chamath Palihapitiya, and Jason Calacanis—spent over an hour mapping how surveillance, censorship, job-loss fearmongering, and a proposed 50% tax on AI companies all converge on a single question: who gets to control the intelligence that will run the economy.
Anthropic released Fable 5 on June 24 to strong benchmark scores but immediate developer backlash. Three policies drew the sharpest fire, each operating with a different mechanism but a common effect: the company decides what you can do with its models, watches everything you type, and reserves the right to give you less than what you paid for.
“I just got downgraded for asking a very simple question,” Calacanis said on the show. “It’s in the database with you now.”
The practical impact hit Friedberg’s agricultural genomics company directly. His team had been using frontier models to design genetic constructs for plant breeding. “Over the last couple of weeks, they’ve begun to restrict the ability to use the models to do that work,” he said. “The premise is that there’s some sort of bioweapon type risk.” His response: move everything to open-source alternatives. Many of those alternatives, he noted, are Chinese.
The panel saw the Fable 5 restrictions not as isolated product decisions but as one piece of a coordinated strategy. While Anthropic was deploying user surveillance and silent downgrading, CEO Dario Amodei published a blog post arguing that “transparency was no longer good enough” and calling for a new regulatory agency—”an FAA or maybe an FDA”—to approve all AI models before release.
Sacks connected the dots directly. “Daario wrote a new blog post saying that transparency was no longer good enough, that we needed to have a new regulatory agency like an FAA or maybe an FDA to approve all models.” He characterized the entire effort as a “sophisticated regulatory capture campaign based on fear-mongering.”
The pattern is straightforward: a company builds a model with surveillance baked into its architecture, then argues that all models need government oversight. The target, as Sacks noted, is unmistakable. “They want to apply that regulation to open source, which is impossible. This would be a great way to scuttle and sandbag open-source models.”
The irony was not lost on the panel. Meta’s Llama, once seen as the obvious counterweight to closed models, was described by Chamath as a strategic failure. “We talked about this two years ago and said Zuck needs to view this through the lens of game theory—take the margin out for everybody else. What a fumble.”
The most immediate consequence of American AI restrictions, the hosts agreed, is that they push developers toward Chinese alternatives. Calacanis put it bluntly: “What are the best open source models today? They’re Chinese.”
Friedberg elaborated on the structural danger. His prediction, stated with high confidence, is that U.S. AI restrictions and political pressure will ultimately benefit Chinese open-source model providers, damaging American economic viability. If American companies cannot use American closed models for sensitive work—because the models censor them or surveil their data—they will use Chinese open-source models instead. “That is a major concern,” he said. “The American open-source models are not as good as the Chinese open-source models.”
The solution, Chamath argued, requires building dedicated compute infrastructure for open-source AI—an undertaking with staggering capital requirements. He laid out the math: a single gigawatt-scale data center costs roughly $100 billion to build. Creating meaningful open-source compute capacity at three gigawatts would require $300 billion. He revealed he is already moving on this, having “put in an offer for another gig in a different place” with the explicit goal of creating “deep liquid access to open source.”
Calacanis added that he controls 2,000 acres of zoned land in Arizona that could support such projects. But the broader question remains: can private capital build open-source fast enough to compete with the closed-model duopoly forming between Anthropic and OpenAI?
Friedberg pointed to a working alternative model. He highlighted how the Collison brothers and other philanthropists funded an open-source genome language model through the Ark Institute. “It’s a good example where a community, in this case the Collisons and others, put money behind it to fund this research and output this open-source model. And I think we see more of that kind of coming down the pipe, which creates a very great advantage against the closed proprietary model ecosystem.”
Chamath framed the governance question in terms every corporate board should be asking. “How do I have control? Who am I allowing to learn off of all of this information? Do I want to have single point of failure risk with respect to AI?”
One reason AI companies can demand regulation with a straight face, the panel argued, is that they have spent years telling the public their technology will destroy employment. Dario Amodei has predicted 50% job loss for entry-level knowledge workers within one to five years. Sam Altman has made similar warnings, though Sacks noted he has recently walked them back.
The data tells a different story. Friedberg rattled off the numbers: the May 2026 jobs report showed 172,000 new positions created, unemployment at 4.3%, and software developer jobs at a three-year high. “The idea that AI is going to destroy jobs is a lite idea that is being disproven every single day,” he said.
His own company requested 15 additional engineering headcount because AI had unlocked new product opportunities. “There are two sides to a business: revenue and costs. On the cost side, AI can reduce some humans, but that effect is nominal. The real opportunity is on the revenue side, where suddenly one engineer can do a hundred times or a thousand times what they used to do. We cannot hire enough people.”
Sacks drew a sharper conclusion about the political utility of the job-loss narrative. “They’ve been out there teaching the public that what they do is harmful for years. How am I supposed to be out there defending it?”
The panel distinguished among the key figures: Elon Musk describes a distant utopia where robots produce so much that work becomes optional. Amodei predicts sharp, near-term job destruction—a claim the data increasingly contradicts. Altman has begun backing away from his earlier predictions.
The job-loss narrative, the hosts argued, has created an opening that Senator Bernie Sanders is now exploiting. On June 20, Sanders published a New York Times op-ed announcing the “American AI Sovereign Wealth Fund Act,” which would impose a one-time 50% tax on the stock—not profits—of the largest U.S. AI companies. The seized shares would go into a government-run sovereign wealth fund, and the public would receive voting rights and board representation at each company.
Sanders’ argument: “The foundation of AI is our collective and human intelligence—the books, songs, journalism, scientific research, code—essentially stolen by some of the wealthiest people in the world.”
Sacks summed up the bind that AI CEOs have created for themselves: “I’m not in favor of Bernie’s proposal because it’s a straight-up confiscation of property. However, I do have sympathy for where it’s coming from—a more voluntary means of allowing the public to participate could work.”
The AI debate unfolded against a backdrop of stubbornly hot inflation. The May Consumer Price Index came in at 4.2% year-over-year, the highest since April 2023, while the Producer Price Index hit 6.5%, a level not seen since late 2022. The European Central Bank hiked rates by a quarter point on June 25, and Polymarket showed a 49% probability of a Federal Reserve rate hike in 2026—a number that had been under 10% before the Iran conflict began.
Chamath analyzed the energy dimension: China has been acting as a global buffer, keeping a lid on oil prices through its controlled consumption. If China’s reserves deplete and it needs to buy an additional 3 million barrels per day, oil could spike past $100 toward $150 or even $200. That would feed directly into CPI and PPI, forcing the Fed’s hand regardless of what AI companies do.
Friedberg tied the inflation problem to its root cause: “The core problem with inflation, all of it, roots back to excess government spending. In order to get elected, you need to give someone something more than they have today.”
The podcast episode captured a hinge moment in the AI industry’s trajectory. The surveillance features in Fable 5 are not bugs—they are architecture. The push for a federal AI approval agency is not hypothetical—it has a named CEO behind it with a published white paper. The migration to Chinese open-source models is not a distant risk—it is already happening, as Friedberg’s own company and countless others make the switch. And the political response, from Sanders’ 50% tax to the Commerce Department’s on-again, off-again export controls on Anthropic’s own Mythos 5, is accelerating faster than most corporate strategies can adapt.
The hosts diverged on solutions but converged on the diagnosis. Sacks predicted the U.S. will “become stuck with only one class of proprietary AI models and a set of opaque rules, while the rest of the world operates without those restrictions.” Friedberg warned that China will “race ahead of the U.S. in biotech, material science, and new industrial systems” if American regulation forces companies onto restricted closed models. Calacanis predicted that Friedberg himself would start building proprietary models on top of open-source foundations—a hybrid approach that may become the default for any company with sensitive data.
For investors, the calculus has shifted. The AI trade is no longer simply about picking winners among U.S. labs. The real question is whether the infrastructure of intelligence will be open or closed, surveilled or private, American-led or Chinese-led—and whether the capital required to build an open alternative can be assembled before the regulatory door slams shut.
References:
Once added, BigGo Finance appears first in Google Search Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
The news and data on this website are for reference only and do not constitute investment advice or an offer to buy or sell. Information is sourced from exchanges and public sources, and may be delayed, interrupted, or updated. While we strive for accuracy, we do not guarantee timeliness, correctness, or completeness. Content may include external links for which we are not responsible. By using this site, you agree that we and our partners are not liable for any losses. Investment carries full responsibility; please carefully assess risks and consult professionals. If there are errors in the content, please contact us for correction.
David Sacks: Anthropic's Fable 5 Surveillance Marks a Regulatory Capture Campaign Disguised as Safety – finance.biggo.com
Leave a Reply