AI Reading for Saturday June 6

NSA is deploying Anthropic's Mythos model for offensive cyber operations; Hegseth's campaign to purge Claude from government systems has largely failed - Gizmodo

Military-industrial complex goes model-agnostic: Trump signed a national security memorandum ordering accelerated AI adoption and AI vendor diversity across US intelligence and warfighting domains. - Reuters

Trump signals interest in the US government taking equity stakes in leading AI companies, describing it as a public partnership. - Bloomberg

Sam-coin: OpenAI and the White House are discussing a government equity stake to seed a public AI wealth fund, CNBC confirmed. - CNBC

if AI can have a public option, maybe other stuff too? 🤔

The US Army warned Congress that proposed Chinese-equipment bans could stall military data center builds because domestic suppliers can't fill the gap - Bloomberg

Beware the 4-fingered drone ranger: Shield AI's $12.7B V-BAT drone program fingered over 50+ crashes, a severed-finger injury, and whistleblower allegations of falsified safety data - Reuters

OpenAI launched Lockdown Mode, restricting ChatGPT network access to block prompt injection attacks but disabling Deep Research and Agent Mode - Engadget

Former OpenAI CTO Mira Murati says OpenAI would have imploded without Sam Altman's return after his 2023 ouster. - Bloomberg

Leaked details suggest OpenAI's GPT-5.6 will improve token efficiency, multimodal capabilities, and cost-effectiveness. - Geeky Gadgets

Clippy gets a call sign: Satya Nadella says AI agents should be governed like employees, with identities, sandboxes, and audit policies, as enterprise agent fleets scale - Business Insider

Microsoft Build 2026 launches Scout autopilots and a sandboxed OpenClaw app, giving enterprise AI agents their own Entra identities and Teams presence - Fast Company

Satya Nadella ‘Not Sure’ Who Said Microsoft Wanted to Make Addictive AI, Is Looking for Guy Who Did This - 404 Media

Waymo has logged hundreds of safety incidents including red-light violations and emergency scene delays, though a viral stunt confirmed its vehicles won't roll over people - Futurism

Waymos seem pretty safe so far, Uber and Cruze programs went defunct after running people over, let’s see what happens to Tesla with no safety drivers.

Google will pay SpaceX $920 million per month to rent 110,000 Nvidia GPUs, in SpaceX's second huge compute deal with an AI rival in weeks - Yahoo

$8k per month seems like a lot. I wonder if Musk retrades the ones they rented to Anthropic in a bidding war.

SpaceX's narrative has expanded from rockets (2002) to Starlink (2019) to AI compute (2025) to projected orbital AI satellites (2028), each layer justifying a higher valuation without the economics having caught up—a pattern analyst Anand Sanwal calls a "Ponzi scheme of ambition." - Fortune

S&P 500 declined SpaceX's request for expedited IPO inclusion, blocking a precedent that could have fast-tracked OpenAI and Anthropic - Ars Technica

Funny money: Michael Burry published a detailed analysis of a deal where Nvidia sold GPUs to an SPV it took equity in, renting GPUs to xAI/SpaceX, funded by debt Apollo stuffed in its captive insurance subsidiary. - Yahoo

DA Davidson analyst says Nvidia's 70%-plus gross margins are safe through 2030 due to hyperscaler dependence on its chips. - Bloomberg

Meta is erecting 125,000-square-foot tent structures powered by behind-the-meter gas turbines, completing AI data center capacity in three months rather than three years - Tom's Hardware

Meta may raise tens of billions in equity to fund AI infrastructure, joining Alphabet in tapping capital markets for the $820 billion buildout - Financial Times

Inflation by inference: AI infrastructure spending is driving up consumer prices in four key categories, compounding inflation pressures tied to the Iran conflict - The Washington Post

Bank of England's Bailey warns energy constraints could force governments to ration AI, with health care and defense competing for priority access - Bloomberg

Marvell cinematic universe Jensanity: Huang named Marvell the next $1 trillion company; shares are up 55% in a week to a forward P/E of 72 - 24/7 Wall St.

BofA strategist Michael Hartnett warns of an AI bubble, noting 21 stocks drove the S&P 500 to records — matching the dot-com peak concentration - Yahoo

AI-boosted markets minted nearly 2 million new millionaires in 2025, lifting global high-net-worth wealth 8.7% to $98.3 trillion - Private Banker International

AI is disrupting wealth management as advisors earning up to $500,000 confront growing chatbot competition. - Bloomberg

Retail investor use of AI for stock-picking rose 75% in a year, though experts warn bots can't replace fundamental market knowledge - Business Insider

AI cited as reason for nearly 40% of May 2026 U.S. job cut announcements, the highest May layoff total since 2020. - CNBC

32% of firms that cut positions citing AI later rehired for the same roles; 55% of employers regret AI-driven layoffs, per Forrester - Fast Company

Tech workers at Amazon, Google, and Apple say AI compresses weeks of work into days, but freed time is immediately consumed by new tasks - Business Insider

AI is enabling larger teams with fewer managers, putting pressure on the "accidental manager" model where promotions serve mainly as retention tools - Fast Company

AI is increasingly used to build bespoke internal tools rather than replace workers, realizing the original vision for personal computers. - The New Yorker

Enterprise AI resembles the internet in 1991: powerful infrastructure exists but the transformative consumable layer equivalent to the web has not yet arrived. - Fast Company

Sergey Brin argues AI will push humans to sharpen their skills rather than become obsolete, citing how Go players improved after AlphaGo defeated them - Business Insider

Economists argue the tax code needs no AI "apocalypse clause" — labor and capital income shares held steady through prior automation waves - Yahoo

AI is more than a light bulb - The Wall Street Journal

a very human-like technology tends to flatten capital-labor tradeoff, make factors more substitutes than complements, in the extreme becoming one factor, and meaning the human can't demand a higher wage than the cost of renting a robot. nobody knows everything.

Companies adopt AI model routing to cut costs, threatening OpenAI and Anthropic's premium pricing model. - CNBC

Power users run local AI models to refine prompts before sending them to ChatGPT, improving output quality and conserving token limits. - Tom's Guide