AI Kills Everybody or Doomer Psyop? OpenAI's Math Breakthrough, Nike's $200B Collapse

The resignation of Jacob Coxon from Anthropic has been framed not as a spontaneous act of whistleblowing but as a coordinated campaign designed to amplify AI doomer narratives and advance a centralized regulatory agenda. The hosts dissect the events around Coxon’s departure, the players involved in rapid message amplification, and the broader historical and political context of such doomer campaigns.

Strategy Behind Coxon's Departure From Anthropic

Coxon, a relatively inexperienced employee with just six weeks at Anthropic and a prior role at an NGO dedicated to halting AI development, resigned and posted a widely shared series of tweets taking aim at the existential risks of AI. Friedberg points out the incongruity: Coxon was elevated as “the expert” after an extremely short industry tenure, despite limited credentials. The public and political figures, including Bernie Sanders, cited Coxon as an authority worthy of deference, underscoring how expertise is manipulatively conferred in the current climate.

Coxon's Account, Inactive Before the Resignation Tweets, Achieved Viral Reach Through Well-Funded Doomer Groups and Political Allies' Amplification

The group discusses the viral nature of Coxon’s tweets, noting that his previously inactive account—with little to no followers and scrubbed prior posts—garnered 110–150 million views practically overnight. David Sacks observes that within minutes of Coxon’s post, three prominent and well-organized doomer groups amplified the message, suggesting a deliberate, coordinated effort. Jason Calacanis underscores how algorithmically improbable such virality is for an anonymous or dormant account without external promotion.

Talen-Funded Advocacy Groups Amplify Coxon's Tweets, Indicating Possible Pre-coordination

Sacks details that the groups pushing Coxon's message—Encode AI, AI Policy Network, and AI Futures Project—are all funded by Jan Talen, a significant mega-donor in the Effective Altruism (EA) community and a Series A co-lead investor in Anthropic. These groups are tightly linked both ideologically and financially with Anthropic, reflecting a networked campaign rather than spontaneous outcry. Sacks also describes how a podcast episode, using the same language about AI existential risk, dropped simultaneously, further indicating planned amplification rather than organic response.

The Wall Street Journal Story Was Published Before the Tweet Storm Began, Indicating the Journalist Received Advance Notice Under Embargo, Requiring Coordination With Coxon and Supporting Groups

A Wall Street Journal article covering Coxon’s resignation went live just minutes before the tweetstorm. Calacanis explains that this is technically known as an embargo and shows pre-briefing; the journalist was alerted in advance so the story could run in synchrony with the social media blitz. This level of preparation points to intricate coordination between Coxon, the advocacy groups, and media partners to maximize impact.

Coxon Resigned After six Weeks at Anthropic, Sparking Questions About His Motives and Canceling His Podcast Appearance to Avoid Direct Questioning

The hosts add that Coxon, aware of difficult questioning, canceled a scheduled podcast appearance, depriving skeptics of the chance to interrogate his process and connections. Mckesson speculates that Coxon may have joined Anthropic specifically to orchestrate this campaign, with his quick exit timed for maximal attention and impact.

Historical Track Record of Doomer Predictions Being Incorrect

Friedberg and Sacks argue that doomer narratives surrounding new technologies, including AI, have a long history of dire predictions that are repeatedly disproven by actual outcomes.

Ai Threats Overestimated: Gpt-2 Release Didn't Lead To Predicted Cyber Risks

Sacks points out that when GPT-2 was released, experts warned it was too dangerous, predicting it would prompt catastrophic cyberattacks and financial system collapse. No such events occurred, and there is consensus that the best defense against AI-enhanced cyber attacks is AI-driven security.

Ai-related Job Losses Overstated; Economy Experiences Job Gains, Tight Labor Markets

Predictions that half of all jobs would be lost or that unemployment would reach 10–15% have proven baseless; in reality, job gains and a tight labor market have occurred.

Ai Capabilities Claims About Danger Are Undermined by Current Systems' Inability to Reliably Complete Tasks Like Hotel Booking or Python Autocomplete

The hosts further note that AI model capabilities are overstated. Everyday tasks—like Python autocomplete or end-to-end hotel room tasks—still require a human in the loop, undermining predictions th ...

The Contradiction Of Claiming ai Risks While Seeking Massive Valuations

Anthropic faces a glaring contradiction in its public positioning. On one hand, leaders and senior employees warn of a greater than 10% risk of their AI causing human extinction by 2030. Lead alignment scientist Evan Hubinger has clearly endorsed these claims, stating publicly that “I personally think it is over 10%” and “We do not yet have a plan to solve alignment.” This isn’t limited to a single disgruntled former employee—numerous insiders have echoed and cosigned these existential warnings.

Yet, at the same time, Anthropic pursues a multi-trillion-dollar IPO, seeking to maximize its valuation amid immense investor demand. This posture invites tension: investors are being asked to underwrite a company whose own experts allege a substantial chance of catastrophic failure. Sacks notes this exposes the company to “the mother of all product liability lawsuits,” as Anthropic continues to release and update its AI models while acknowledging their unsolved safety.

These risk admissions have a direct financial impact: investors, alarmed by existential risk claims, demand significant discounts or refuse to participate, undercutting any premium IPO ambitions. Meanwhile, Anthropic attempts to square the circle with self-aggrandizing claims—asserting that only they can safely develop frontier AI and calling for regulatory structures that enshrine a monopoly or duopoly. As Sacks and Calacanis emphasize, these claims lack credibility in the broader market, since investors and observers recognize the proposal as nakedly self-serving and unsustainable for public scrutiny.

Ipo Legal and Disclosure Issues From Internal Safety Claims

The legal and regulatory stakes are severe. Hubinger’s public endorsement of extinction-level risk transforms what could have been dismissed as disgruntled whistleblower “vibes” into official pre-IPO company positions. With an S-1 filing in process, Anthropic’s executives are exposed: the statements are material, and uncoordinated communication from safety leads and employees may constitute violations.

Chamath explains the dilemma: in the S-1, Anthropic must either openly acknowledge massive product liability risks—which would greatly depress valuation—or disavow the employee and executive claims, risking a revolt among core teams who genuinely believe in the existential danger of their work. Quiet periods demand tight message discipline and centralized communication. But at Anthropic, internal alignment has broken down, and the disclosed risk claims have reached a scale far beyond normal pre-IPO risk factors. Typical S-1 risk sections discuss regulatory, competitive, or supply threats—not existential risk to civilization verified by company insiders.

This scenario resembles other infamous IPO legal headaches, such as Slack’s network effects comments or the Google Playboy interview, but the Anthropic disclosures are far more destabilizing. The SEC’s pressure-testing of S-1 filings is intended precisely to surface these kinds of material issues. Now, the regulators and lawyers must grapple with whether Anthropic’s public risk assertions are properly disclosed—and whether these are compatible with listing the company at all.

Liability and Market Consequences Of Acknowledging ...

David Friedberg and David Sacks express serious concerns about the trajectory of federal AI regulation in the U.S., arguing that it risks creating a centralized system of control that suffocates open-source innovation and dangerously consolidates power over critical technologies and information.

Proposed Fda Ai Framework Enforces Duopoly Favoring Frontier Labs

Friedberg highlights that federal AI regulations, as proposed, would require extensive approval processes and safety certifications. These are navigable only by well-funded, large corporations—referred to as “frontier labs”—who can afford the corporate and compliance infrastructure required to satisfy these new standards. Open-source projects, which are developed by decentralized communities rather than formal companies, lack the capacity to navigate this regulatory environment. As a result, these rules would effectively exclude open-source AI from the marketplace.

Heading: Open-Source Ban Increases Frontier Companies' Market Control

Friedberg warns that as regulatory burdens increase, open-source AI—unable to comply—may be banned outright. This would hand even more market control to the few big companies working closely with federal regulators, creating a de facto AI oligopoly. He likens this to a situation where, if open-source is excluded, the U.S. might fall dramatically behind in global technological development, becoming isolated while other regions advance.

Officials Gain Control Over Ai Progress "Gas Pedal"

Friedberg emphasizes that regulation centralizes a “gas pedal” for AI progress, with officials granted the authority to slow down, halt, or selectively accelerate AI development. He describes this as “a system of control” where power is given to a select regulatory body or new federal agency, all centralizing decision-making into the hands of a small group of individuals. This dynamic means open-source projects could be perpetually stalled or blocked at will.

Dangers of Centralized Ai On Critical Systems and Data

David Sacks draws parallels to the trust and safety regime implemented during the COVID-19 pandemic, in which government agencies pressured social media companies to suppress dissenting viewpoints and define “disinformation.” He warns that, under a federal AI department, similar dynamics will play out not just on social media but inside AI models themselves, particularly in sensitive domains like health.

Ai Under Government Content Policies Would Offer Only "Official" Health Information, Compromising Autonomy and Decision-Making

Sacks points out that with a federal AI authority setting the official standards for AI outputs, AI models will only be allowed to present approved information, such as government stances on vaccines or treatments. Users would no longer receive diverse perspectives or critical analysis; instead, all answers are filtered through the government’s definition of safety and truth, compromising user autonomy and informed decision-making.

Government Control Suppresses Dissent and Alternative Viewpoints

Sacks notes that the suppression of dissenting or alternative viewpoints on vital issues—practiced on social media during the pandemic—would become far more comprehensive and automated under a centralized AI regime. AI could be forced to automatically reject or censor opinions or analyses that depart from official narratives, effectively silencing debate and curtailing personal agency.

Federal Ai Agency to Regulate Ai Development, Developers, and Outputs, Centralizing Information Authority

Both Friedberg and Sacks warn that a new federal agency (the “FDA AI") would be empowered not only to regulate AI development but also to dictate who can develop AI, what can be developed, and what outputs AI is allowed to generate. This agency would thus become the arbiter of truth and permissible discourse across all AI systems—placing an unprecedented degree of informational and technological control in the government’s hands.

Importance Of Preserving Open-Source Ai to Avoid Centralized Control

Friedberg argues that open-source AI is essential to democratizing the benefits of advanced technology. Unlike proprietary models, open-source eliminates the need for expensive data centers and centralized cloud providers, allowing AI to run locally on individual machines. This dramatically reduces cost—by as much as 50 times, according to Friedberg—and allows anyone to benefit from AI’s productivity enhancement ...

AI services are revolutionizing how individuals and organizations work, but they bring significant risks to the privacy and security of sensitive communications and intellectual property (IP). Leading industry voices express urgent concerns about current legal protections, the genuine possibility of IP leakage, and the unfair market advantages accruing to providers of frontier AI models. In response, organizations are rethinking their approach to using AI, pushing toward sovereign infrastructure and open-source options to safeguard their data.

Insufficient Legal Protections For AI Chat Data vs. Email and Attorney-Client Communications

Current data privacy laws provide weaker protection for AI conversation data than for traditional forms of digital communication like email. David Sacks highlights that emails are protected by laws requiring the government to obtain a search warrant and prove probable cause in court before accessing content. In contrast, AI chat data usually requires only a subpoena or simple court order for government access. This lower threshold is concerning as AI becomes a substitute for professional advisors, such as doctors, therapists, and lawyers.

Further exacerbating the privacy risk, conversations with AI do not enjoy privilege protections provided to communications with licensed professionals—such as attorney-client privilege. If someone asks a sensitive legal or medical question to an AI, that interaction is not confidential and is not shielded from legal discovery like a conversation with a human expert would be. David Sacks asserts that data privacy rules must evolve to reflect how people now use AI tools for personal and sensitive matters, as individuals cannot currently depend on confidentiality when discussing personal or even proprietary issues with an AI.

Another significant risk lies in the training and improvement of frontier AI models. Frontier AI providers, like OpenAI, state they may use de-identified user data for further training of models, leaving an open possibility for user communications to be incorporated into the systems powering these AI platforms. While companies claim personal information is stripped, the reality is that insights or approaches derived from a user's sensitive or proprietary data—like scientific research or novel business strategies—can inform future model capabilities, bypassing direct IP protections.

How De-identified Data Extraction Enables Competing With Customers

This practice of extracting de-identified data creates a slippery slope: insights and approaches originating from proprietary conversations may be used by AI providers to enhance their models, enabling those same providers to enter vertical markets and build competing products. David Friedberg describes firsthand experiences where confidential, novel scientific ideation shared with an AI model appears to resurface later in the same model's responses, despite there being no new publicly available source for that information—suggesting leakage of core organizational IP.

De-identification—removing explicit personal or company identifiers from the data—does not prevent the extraction of valuable problem-solving approaches. AI models can generalize methods and strategies harvested from users’ iterative chats, which then improve the model’s performance in solving similar problems for other customers or for the AI provider itself. This creates a potent "network effect," with closed AI platforms aggregating collective user intelligence and embedding it into their models, thus granting themselves an informational and competitive advantage unavailable to individual users or companies.

OpenAI, Anthropic, and other leading AI companies have already displayed willingness to compete in vertical market applications, developing products that rival those of their own customers who trusted the platforms with sensitive data. The example of Anthropic’s Claude product competing directly with a major customer, Cursor, after their data was likely leveraged, demonstrates the risk. As long as these AI providers retain the right to use aggregated, de-identified data, customers expose themselves to unfair competition and the potential loss of unique IP—unless strong privacy laws or structural changes are enacted.

Market Response and Solutions For Protecting Proprietary Information

As the scale and awareness of AI-driven data leakage grows, enterprise customers are moving away from relying solely on API-based or SaaS AI offerings from major frontier model providers. Companies are increasingly demanding sovereign AI infrastructure, deploying models on ...

Performance to Narrative Identity Shift Caused Brand Erosion

Nike’s initial dominance in the U.S. athletic shoe market stemmed from its powerful alignment with elite athletes and mastery in sports. Jason Calacanis notes that by 1980, Nike held a 50% share of the U.S. athletic shoe market and revolutionized the industry with innovations like the Air Jordans in 1985 and the “Just Do It” campaign in 1988. David Sacks describes Nike as a brand synonymous with victory, built by legends such as Michael Jordan, Tiger Woods, Serena Williams, and Pete Sampras. Chamath Palihapitiya recalls growing up seeing Nike’s North Star as the embodiment of mastery and excellence through top athletes, making the brand aspirational and inspiring consumers to attain a higher level of personal achievement through association with these figures.

However, Nike shifted away from its foundation of excellence and moved toward activist messaging, privileging political narratives over athletic achievement. This shift was epitomized by major campaigns, including featuring Colin Kaepernick, whose activism took center stage instead of his athletic prowess, and the partnership with Dylan Mulvaney, bringing non-athletes into the brand image. Sacks and Calacanis argue that these campaigns showed Nike cared more about appealing to political or “woke” trends rather than maintaining the excellence and aspiration that originally defined the brand. The pivotal Kaepernick campaign and subsequent partnerships overrode Nike’s previous focus on athletes who were masters of their craft, diluting the aspirational value. Sacks further criticizes Nike for promoting narratives disconnected from its $51 billion global brand positioning, and notes Nike’s stock fell 80% from its peak, with $200 billion in value erased.

The Operational Mistakes Compounding the Branding Failure

Nike’s operational changes further undermined its performance-based legacy. In 2020, under CEO John Donahoe, Nike adopted an aggressive direct-to-consumer strategy, cutting ties with retail partners who had traditionally supported the brand's presence. Calacanis observes that ending relationships with these partners inadvertently opened space in the market for brands like On Running and Hoka to expand. As Nike stores closed, competitors like On—with Roger Federer as the embodiment of modern mastery—moved in to fill the void.

Nike also restructured its organization, dissolving specialized divisional expertise in specific sports—for example, basketball, football, and tennis—replacing them with generalized men’s, women’s, and kids’ categories. Sacks criticizes this reorganization, questioning why Nike would break what was working and lose critical product know-how.

Product quality also declined, betraying Nike’s core brand promise. Friedberg notes that Nike shoes became flimsy, wearing out in six weeks, which contrasted sharply with their previous reputation for high durability and performance. Palihapitiya and Friedberg both recount personal experiences of abandoning Nike in favor of On and Brooks because of this decline in quality and alignment.

The Competitive Opportunities That Emerg ...