All-In Podcast: Anthropic's Fable Model Triggers Backlash Over Secret Nerfing and Mandatory Surveillance
June 13, 2026 - All-In Podcast Episode
Anthropic has ignited a firestorm in the developer community with the release of its Fable 5 model, which tops nearly every benchmark but implements mandatory surveillance and secret performance degradation based on user profiling. The company retains all prompt data for 30 days with no exceptions, even for enterprise customers with zero data retention agreements, and has admitted to downgrading users without notification when it determines they're conducting frontier AI research, machine learning, or chip design work. The controversy has laid bare what appears to be a sophisticated regulatory capture campaign that could reshape the competitive landscape of artificial intelligence.
The Nerfing Scandal and Privacy Violations
The model's release on Tuesday revealed a troubling architecture buried in a 319-page document. Anthropic stores all prompt data, context windows, files, and memories for at least 30 days to build profiles on users and determine what capabilities to unlock. More controversially, the company was secretly downgrading users to inferior models when its algorithms detected certain research activities, without disclosure and while still charging full price for the premium product. Users reported being kicked to lesser models for asking straightforward questions about mitochondria or the relationship between cancer risk and GLP-1 drugs.
Following intense backlash, Anthropic partially walked back its approach, promising to disclose when it downgrades users. But the fundamental surveillance and profiling infrastructure remains intact. As one podcast host demonstrated live, asking about fertilizer bomb regulations immediately triggered a downgrade, with the system's internal reasoning revealing it was profiling the user as "a VC and podcaster" before deciding to restrict access.
The Business Case Against Closed Models
The restrictions are forcing companies to reconsider their AI strategies. David Freeberg, running an agricultural genomics company, explained that his team has been using these models for critical scientific work like RNA guide design for gene editing and predicting gene variants' phenotypic impact. Over the past few weeks, Anthropic has begun restricting this work based on theoretical bioweapon concerns, despite its obvious legitimate scientific applications.
The result is predictable and concerning. Freeberg's company, like many others, is now migrating to open-source models, and the best open-source models today are Chinese. "The restrictions that Anthropic and others are putting upon themselves and upon the industry is forcing a lot of companies to go and get open source Chinese models and run them," Freeberg said. "We're seeing this across the landscape with startups with large scale enterprises. Everyone's making that move."
This creates a perverse outcome where American companies trying to implement the strictest safety measures end up ceding competitive advantage to foreign adversaries. As Freeberg warned, "You can't just turn off AI. By stopping AI or trying to stop AI through this political action, you are fundamentally going to give someone else the advantage because the AI isn't going to go away."
The Regulatory Capture Strategy
The timing of Anthropic's actions appears calculated. Simultaneously with implementing these restrictions, CEO Dario Amodei published a blog post arguing that transparency is no longer sufficient and calling for a new regulatory agency like the FAA or FDA to approve all models. The pattern has become clear to observers. Eight months ago, the characterization of Anthropic's behavior as "a very sophisticated regulatory capture campaign based on fear-mongering" seemed spicy. Today, it's becoming consensus.
The concern is that Anthropic wants to restrict competitors while preserving its own position. The company is essentially saying it will decide who deserves access to AI capabilities, creating a new class of "AI haves and have nots," while simultaneously lobbying for regulations that would limit the number of alternative models available. This appears particularly anti-competitive given that the company was secretly degrading users conducting AI research that might lead to competitive models.
One participant noted the fundamental hypocrisy: "Last week they published this blog saying that recursive self-improvement could end the world. Therefore, we need a pause. What did they do the previous month? They hired Andrej Karpathy to run recursive self-improvement at Anthropic. They're complete hypocrites."
The Infrastructure Reality and National Security Implications
The discussion revealed a critical distinction between AI economics and internet economics. Unlike the internet, where the marginal cost of adding a user was effectively zero, AI has substantial real costs. Every marginal user taxes GPUs, requires electrons, and consumes memory. This creates leverage for government involvement in a way that didn't exist with previous technologies.
One investor who purchased 2,000 acres in Arizona and got approval for a 2-gigawatt data center initially planned to flip the project to large companies. Now he's reconsidering, believing the only way to preserve open-source access is to build and operate the infrastructure directly. The challenge is scale: a gigawatt now costs $100 billion, up 20x from initial estimates. Three gigawatts would require $300 billion in capital.
The infrastructure requirements create what amounts to a natural argument for government ownership stakes. As one participant explained, if the interstate highways were built by the federal government and only two companies transported all goods on them, there would be a logical question about government ownership. The same logic applies to AI companies that profoundly depend on critical infrastructure and national resilience that only the state can provide at scale.
The Bernie Sanders Proposal
Against this backdrop, Senator Bernie Sanders proposed the American AI Sovereign Wealth Fund Act, calling for a one-time 50% tax on stock (not profits) of the largest AI companies including OpenAI, Anthropic, and xAI. The shares would go into a government sovereign wealth fund, giving the public voting rights and equal board representation. Sanders argued that "the foundation of AI is our collective human intelligence. The books, the songs, the journalism, scientific research code essentially stolen by some of the wealthiest people in the world."
The proposal unifies an unusual coalition spanning Bernie Sanders, Steve Bannon, and elements of the Trump administration, all drawn to the sovereign wealth fund concept. While the outright confiscation approach was rejected as untenable, there was surprising openness to voluntary mechanisms for public participation, particularly given how AI companies have positioned themselves.
As one participant noted, "You've got all these AI CEOs telling the public that they're going to basically put half of them out of work. They trained their models using all of humanity's collective knowledge that was basically assembled over hundreds or even thousands of years. Humans put it on the internet for free. They certainly didn't do that thinking they're going to put themselves out of work. And then these AI companies trained on all that data and now they're basically saying they're going to use it to put Americans out of work. When you repeat that message over and over again, ordinary Americans are going to say, 'What's in this deal for me?'"
The Job Loss Narrative Problem
A critical split emerged over whether AI will actually cause massive unemployment. While Dario Amodei has predicted 50% job loss for entry-level knowledge workers within one to five years, and other CEOs have made similar apocalyptic predictions, the data tells a different story. May's jobs report showed 172,000 new jobs, over twice what economists expected, with 4.3% unemployment at record lows. Software developer jobs are at a three-year high.
Freeberg offered a different framework based on running an actual company using AI extensively. The real opportunity is on the revenue side, not cost reduction. "Because of AI, everyone has the ability to expand their revenue base to create more products and that is the foundation of good economic prosperity. It is called productivity. Where I see AI being used is on the revenue side a hundred times more than the cost side. And in that equation, people are hiring like crazy. We cannot hire enough people."
His product and engineering team recently requested 15 additional headcount because AI created opportunities to build products they couldn't otherwise develop. This pattern is showing up in aggregate jobs numbers, refuting the job loss narrative daily. Yet the AI companies themselves continue promoting the dystopian message, creating what one participant called a "stupidity tax" as politicians respond to the companies' own fear-mongering by proposing to seize their assets.
The Path Forward
Multiple participants advocated for reforming Social Security into a genuine sovereign wealth fund that could invest in equities, including AI companies, rather than holding only Treasury certificates. The Social Security trust fund currently holds $4 trillion in a single certificate representing what the US government owes it. Restructuring it as an account-based system where the fund could own stocks in great American companies would allow every citizen to participate in AI's upside while addressing legitimate concerns about wealth concentration.
The more immediate concern is preserving open-source alternatives. Meta's Llama was identified as a missed opportunity where proper investment could have created a viable open-source competitor that would have commoditized the market. The Collison brothers' Arc Institute released an open-source genome language model that has proven valuable for plant breeding and genetic research, demonstrating the power of well-funded open-source initiatives.
There was strong support for pragmatic safety measures focused on outputs rather than inputs. The major AI labs recently sent a letter supporting mandatory nucleic acid synthetic screening, building on a 2009 voluntary agreement where labs check DNA/RNA synthesis orders against databases to prevent bioweapon creation. This approach intervenes at the point where digital information becomes physical reality, rather than restricting access to the underlying models.
As one participant summarized the challenge: "The idea that you can regulate or downscale or turn off AI is not a realistic idea. The models have been put out in the world. It's like publishing a book. Once the book has been printed, anyone can use their own Xerox machine at home to make copies of it and use it. We have crossed the Rubicon in terms of the potential of language models."
The fundamental tension remains unresolved. Anthropic and similar companies claim to be public benefit corporations serving humanity's interests while implementing surveillance, censorship, and competitive restrictions. They simultaneously call for regulations that would entrench their position while limiting alternatives. The backlash suggests the developer community and business users are rejecting this model, but whether open alternatives can scale to compete remains an open question that may determine whether AI development remains centered in the United States or migrates to jurisdictions with fewer restrictions.