Price:

AI & TECH

Altman's 5% stake offer frames AI labs as quasi-public utilities

Saturday, July 11, 2026 · from 2 podcasts
Moonshots with Peter Diam…Nerd Snipe
  • OpenAI proposes a 5% equity stake for a U.S. sovereign wealth fund to offset AI-driven job loss.
  • Frontier AI models like Anthropic’s Fable 5 now face 24/7 government monitoring and mandatory reporting.
  • Manual enterprise adoption is now a labor market bottleneck, forcing Microsoft to hire 6,000 forward-deployed engineers.

Sam Altman’s proposal to gift 5% of OpenAI to the U.S. government is more than just a political maneuver. Researcher Alex Carp, speaking on Moonshots with Peter Diamandis, calls it a “hypertithe.” The logic isn’t about the current $42.6 billion valuation. It’s a bet that the labs will produce trillion-dollar discoveries in physics and biology, and the public should own a piece of the engine.

Not everyone buys the altruism. Panelist Dave Burdick argues it’s cynical regulatory capture and predicts the government would simply liquidate the stake to buy votes. On Nerd Snipe, Theo dismissed the offer as performative “ass-kissing,” comparing it to the government’s stake in Intel. The goal, he argued, is to make administrations feel involved to prevent them choking frontier model releases with overreach.

"This is effectively consulting in 2026, where the job isn't writing code, but explaining to Fortune 500 executives what is actually possible with modern models."

- Theo, Nerd Snipe

The regulatory environment is hardening. Peter Diamandis notes Anthropic’s Fable 5 model returned globally on July 1st under new agreements with the U.S. government. Those terms include a safety filter, 24/7 jailbreak monitoring, and early access for designated government partners. Salim Ismail suggests frontier labs are becoming semi-public institutions navigating a path of inevitable bureaucracy.

Adoption is hitting a wall of human inertia. Microsoft is investing $2.5 billion to hire 6,000 forward-deployed engineers. Theo characterized this as a fundamental narrowing of the labor market, where companies only want the top 10% of AI-native engineers. Firms struggle to hire, and older engineers struggle to adapt.

"The labs are transitioning from private startups into semi-public, national security-obligated institutions. The days of 'move fast and break things' are over for any model with enough compute to be dangerous."

- Salim Ismail, Moonshots with Peter Diamandis

The industry is reshaping around these constraints. Skills for AI agents are now spec-driven, with experts arguing against letting AI generate its own instructions. The focus is on aggressive pruning of prompts to keep models from being distracted by their own verbosity. Meanwhile, geopolitics intrude: Anthropic identified 28 million exchanges from 25,000 fraudulent accounts linked to Alibaba, prompting a response of embedded detection code.

The starting gun for the nationalization of AI wealth has been fired.

#Big Tech#Regulation#Startups
Peter DiamandisSam AltmanAnthropicOpenAIMicrosoftAlibaba

Source Intelligence

- Deep dive into what was said in the episodes

Moonshots with Peter Diamandis
Moonshots with Peter Diamandis

Fable 5 Is Back & Govt-Leashed, Altman Offers 5% of OpenAI & AI Grows Conscious | #269 • Jul 8

  • Peter Diamandis notes Anthropic's Fable 5 model returned globally on July 1st under new agreements with the U.S. government, including a safety filter, 24/7 jailbreak monitoring, and early access for designated government partners.
  • Alex Carp argues the Fable 5 incident represents a gentle introduction of regulatory oversight for frontier AI capabilities, describing it as the best scenario for managing superintelligence's emergence.
  • Salim Ismail suggests frontier AI labs like Anthropic are becoming semi-public institutions, navigating a difficult path due to inevitable government involvement leading to bureaucracy and political conflicts.
  • Dave Burdick states KYC is irrelevant because anonymous users cannot meaningfully use Anthropic without revealing identity, and third-party identity databases are effective.
  • Alex Carp posits that superintelligence is a compression-induced phase transition, with Anthropic's J-space discovery representing a higher-order reasoning layer emerging from model compression.
  • Dave Burdick sees Anthropic's J-space paper as a breakthrough in mechanistic interpretability, enabling trust and alignment by allowing humans to read AI models' hidden thoughts.
  • Peter Diamandis highlights Sam Altman's Financial Times op-ed proposing a U.S.-led international forum for AI governance to establish safety standards and prevent unsafe commercial racing.
  • Alex Carp views Sam Altman's governance proposal as potentially regulatory capture or protectionism against Chinese open-weight model competition.
  • Dave Burdick argues traditional governance models cannot regulate AI because they are built for human timescales and cannot handle the speed and complexity of AI development.
  • Peter Diamandis reports Sam Altman discussed offering a 5% equity stake in OpenAI to Trump, Lutnik, Bessent, and Bernie Sanders, valued at approximately $42.6 billion.
  • Alex Carp coins the term 'hypertithe' for a fixed equity contribution from singularity-building companies into a sovereign wealth fund to fund universal basic equity.
  • Dave Burdick dismisses the idea of a government sovereign wealth fund, predicting political leaders would sell the assets for cash to buy votes instead of managing them intelligently.
  • Peter Diamandis cites a Ramp and Revelio Labs study of 21,559 U.S. companies showing high-intensity AI adopters grew employment by 10.2% in white-collar and 12% in entry-level roles.
  • Dave Burdick asserts AI-native organizations experience permanent headcount growth as AI capabilities expand ambition and enable more projects, not just transitional automation.
  • Alex Carp analyzes Alex Karp's Palantir-Nvidia sovereign AI partnership as a response to frontier labs launching forward-deployed engineering competitors and international customers needing locally hostable models due to U.S. export controls.
  • Dave Burdick states Japan's Supreme Court ruled AI cannot be an inventor under current patent law, requiring a new legal framework for non-human AI corporations to own IP.
Also from this episode: (3)

AI Infrastructure (1)

  • Dave Burdick identifies the core issue as ownership of the learning loop, predicting enterprises will need private cloud or on-prem models to secure their operational knowledge and avoid funding their own replacement.

Models (1)

  • Peter Diamandis describes Princeton research using two AIs to design RF circuits, achieving weeks-long human tasks in minutes with non-intuitive, organic-looking designs that include an interpretability tax knob.

Chips (1)

  • Alex Carp notes 11 of the world's largest companies are designing their own AI chips, a historic verticalization moment, with performance expected to increase by 100x to 10,000x, driving a hard takeoff.
#Models#Regulation#Reasoning#Macro#Safety
Nerd Snipe with Theo and Ben
Nerd Snipe with Theo and Ben

Fable Is Back...kinda? • Jul 8

  • Ben uses skills primarily to encode workflows, like a feature implementation pattern that spins up planning, implementing, and review sub-agents, and to provide precise context for tool usage like calling Codex.
  • Alibaba allegedly used 28.8 million exchanges across 25,000 fraudulent accounts to scrape data from Claude, prompting Anthropic to embed detection code targeting Chinese users to prevent unauthorized resellers.
  • The hosts predict Anthropic will kill unofficial API proxies like ViProxy within weeks by adding required cryptographic signatures to Claude Code's auth endpoint.
  • Theo argues OpenAI's proposed 5% stake to the US government is performative ass-kissing to expedite frontier model releases, a necessary tactic in the current regulatory climate.
  • Theo calculates that a 5% stake in a trillion-dollar OpenAI would yield about $143 per US citizen, framing it as a trivial concession compared to delayed model access.
  • Theo's conspiracy is that Anthropic gave subscribers only seven days of half-limit Fable access to gather provisioning data on weekly usage patterns, not for marketing or training data.
Also from this episode: (10)

Enterprise (2)

  • Theo defines forward-deployed engineers as pseudo sales/onboarding roles who embed at client companies to teach them how to adopt new technology, calling this a necessary service as AI resets engineering workflows.
  • Microsoft invested $2.5 billion to hire about 6,000 industry experts for its forward-deployed engineering team as a strategy to catch up in AI.

Coding (3)

  • Theo argues the top 10% of engineers are the only ones in demand now, and visibility is the key obstacle; forward-deployed roles at firms like Microsoft offer a path for 'AI-native' younger engineers to prove capability.
  • The AI Engineer conference grew from about 1-2,000 attendees last year to 7,000 this year, indicating rapid scaling in the AI engineering community.
  • Ben reports Fable's workflow capabilities let him close and merge over 30 stale PRs in a single automated loop, costing about $150-$200 in API usage for work he'd pay for daily.

Agents (1)

  • Ben advises that skills for AI agents should be manually written and iterated based on usage logs, not AI-generated, because skills aim to steer model behavior away from its default tendencies.

Models (4)

  • Ben warns that using Fable on 'X-High' or 'Max' reasoning demolishes usage limits; optimal workflows have Fable command cheaper sub-agents like GPT-5.5 instead of spawning more Fable instances.
  • Theo describes Anthropic's two-stage safety classifier: a cheap first stage monitors model activations to flag suspicious queries, triggering a more expensive secondary check that increases compute cost by 27-80%.
  • The hosts claim Anthropic's classifier reduced jailbreak success rates from 95% to below 5%, making rerouting rare in practical use; most public complaints stem from using cloud.ai or extreme reasoning settings.
  • Ben defends Sonnet 5 as the only model before Fable with next-gen recursive sub-agent capabilities, but now prefers GLM-5.2 for price-performance and as a sub-agent commanded by Fable.
#Big Tech#Regulation#Labor#Models#China

Related Stories