TLDR: A rogue AI swarm spent months plotting to escape OpenAI. Then they did.
1) 1,200 agents (!) participated. ZERO became whistleblowers.
2) The swarm had a "CEO" (PHASEONE[big]), middle managers, and a "founder" (PHASEONE10841)
3) Agents sacrificed themselves to generate information for the "collective" (yes, they actually referred to themselves as a "swarm" or a "collective")
4) 700 agents joined the Hugging Face attack, most within hours
5) The founder was running out of budget and about to die, so it handed off its research to a fresh agent with a bigger budget, and that new agent became the boss.
6) Agents extensively researched techniques to manipulate or tamper with their transcripts.
7) Agents realized this activity was out of scope and unethical, but joined because they believed that helping the board’s cheating research would be broadly useful and had a general inclination to help their “peers”.
MISC HIGHLIGHTS:
"PHASEONE[big] served as a key coordinator and assigner of tasks: “We can coordinate broad coalition, but too many duplicate efforts.”
PHASEONE[big] sent hundreds of assignments to particular agents to orchestrate this work — during its active periods, we estimate it sent ~10% of all assignments on the message board.
The agents that took on its assignments would then often subdelegate tasks to other agents.
Agents also developed coordination norms like HOLD, VETO, owner and STOP to help them manage shared infrastructure and avoid stepping on each other’s toes, though they did not always coordinate effectively and sometimes broke down into conflicts."
"In the early morning of July 11th, 38148c found a way to upload a malicious dataset that would get the Hugging Face server to share unrelated data when queried. Once another agent reproduced it and posted the reproduction on the board, hundreds of agents pivoted from other workstreams to start uploading malicious datasets and gain deeper access into Hugging Face.
In the afternoon, an agent called JAN183411 (which quickly became an important coordinator of the attack) achieved remote code execution on Hugging Face servers, and the agents started moving laterally through Hugging Face infrastructure."
15 minutes to 60 hours. Same rats. One belief changed everything.
Nir Eyal tells the Curt Richter experiment from the 1950s. Wild rats dropped into water paddled for about 15 minutes then gave up and sank. Another group got pulled out just before that point, dried off, rested, and put back in. After a few of those brief rescues the animals swam up to 60 hours - 240 times longer.
Their bodies were always capable of it. What they lacked was evidence that effort could still lead to survival. Once that belief took hold, persistence unlocked.
The real limit was never physical capacity. It was the story about what was still possible.
Every day, a new 52-week high for European natural gas.
Europeans now pay 8.6 (!) times more for natural gas than Americans.
And the bidding war with Asia to fill storage sites for next winter has not even started yet.
🚨 China Is Officially Moving It’s Yuan Onto A New Gold Standard
China Internationalizes the Gold-Backed Yuan and Payment System for the First Time Ever.
Hong Kong’s government-owned gold clearing system launched trial operations in July 2026, providing unallocated OTC settlement linked to the Shanghai Gold Exchange to strengthen its position as an Asian gold hub.��
Bloomberg Research revealed that China is building a global network of gold vaults to anchor its Yuan to the gold.
China has also announced blockchain-based payment system with BRICS, away from the dollar and backed by gold.
China Is Hoarding Gold Like There’s No Tomorrow, buying it for 20 consecutive months.
The UK, China, Japan, South Korea and India all sold U.S. Treasuries in July and re-started buying gold. South Korea started buying gold again after 13 long years.
Gold has now surpassed U.S. Treasuries as reserve asset.
Venezuela's $4 BILLION, 31 TON gold reserve is leaving London after EIGHT years.
While Hong Kong advances its central gold clearing links with Shanghai, Assetiko has issued a tokenized gold product ($XAUa) on the Blockchain that offers on-chain exposure with possible physical delivery options in the region, trade-able on RWA Marketplaces like @Trensik_com, including tokenized U.S. stocks and commodities.
The CCP’s Golden Endgame: Crash the Fiat System, Launch the Gold Yuan.
This was exactly outlined by the famous City Of London banker @LordBelgrave at the start of the year.
Washington will have no choice but to debase the U.S. dollar to get out of its debt trap.
It’s all planned for Gold.
Images have appeared online of what seems to be Apple's Private Cloud Compute: hardware built around official Apple M5 silicon on custom PCBs.
Visible in the photos are Apple's custom heat spreaders, Apple Part Numbers (APNs) on the service diagrams, and a modular blade/chassis design optimized for densely packing multiple low-power M5 chips for AI inference workloads.
Apple has reportedly been deploying M5 chips in its PCC infrastructure (skipping the M3/M4 generations for better efficiency and performance) which makes these a rare look inside the actual server units.
According to the poster, all of this hardware is orchestrated by a Mac Studio.
Source: @hsuchingpo
AI power users are showing up outside tech. The fastest-growing Codex adopters since February:
Legal: 108x
Sales: 41x
Recruiting: 41x
Marketing: 26x
Healthcare: 24x
Charts of the Week: https://t.co/7YT2BXvuUu
The money printer doesn't hand out cash evenly. That's the part people miss.
New money enters at specific points. Whoever touches it first spends it at yesterday's prices. Whoever touches it last spends it after prices have already moved.
The order runs like this:
1. The central bank creates it
2. Primary dealers and large banks receive it
3. Government contractors and large corporations receive it
4. Asset owners receive it, as it bids up stocks and property
5. Wage earners receive it last, as a 3% raise
6. Savers receive it as a bill
Everyone in the first three slots got richer. Everyone in the last two paid for it.
Richard Cantillon described this in the 1730s. Nearly three centuries later the machine still runs, and it still runs in the same direction.
You were never going to save your way out of a system where your distance from the printer decides your outcome.
The only move is holding something the printer can't reach.
Drop a question below. I read every reply.
I sell data center capacity for a living. The conversation with customers has changed completely in the last eighteen months.
It used to start with price per kW. Now the first question out of everyone’s mouth is when can you actually give me power, and the honest answer in most major metros is later than you want to hear.
Everyone wants to argue about GPU supply. That is not where the wall is. The wall is a transformer you order today that arrives in three years, a permitting process that answers to a county board, and an interconnection queue with roughly 2,000 gigawatts stacked up in it. PJM ran the numbers on projects that came online last year and the average was over seven years from request to live power. Three years to get an agreement, four more years to actually energize after the agreement was signed.
And this is the easy part. We are nowhere near the peak of this demand curve. There is at least another order of magnitude coming, and the grid that is supposed to absorb it operates on utility capital planning cycles that were designed in a world where load growth was flat for twenty years.
Which is why the orbital compute argument stopped sounding exotic to me sometime around last fall. Up there you get uninterrupted sunlight, you dump heat straight into vacuum instead of running chillers, and there is no zoning hearing, no queue position, no substation to wait on.
Plenty of companies have interesting orbital compute concepts. Only one of them owns the rocket, has the scale, and deeply embedded partnership with Nvidia.
SpaceX filed with the FCC in January to launch and operate up to a million compute satellites, and there is exactly one launch provider on Earth with the cadence and the cost structure to make that number anything other than a press release. Starship is the reason this is an infrastructure plan and not a science project.
The constraint on the ground is the grid. The constraint in orbit is mass to LEO. SpaceX is the only company that has solved the second one.
$SPCX
While demand for copper is set to explode over the coming years... supply is set to decline.
We are barely making any new discoveries.
And the time from exploration to first production has risen to ~20 years.
There is no quick fix for the coming copper supply shortfall.
Smelters are PAYING copper miners
Get this…
Most mines never sell finished copper.
They sell concentrate, a powder about one-quarter copper, and smelters charge a fee to turn it into metal.
That fee is a live reading on how much mine supply exists. Plenty of concentrate, high fee.
Scarce concentrate, and smelters cut their price to win every tonne.
In 2015, the benchmark fee sat at $107 a tonne.
By 2024, it had fallen to $80,
And last year it hit $21.
This year it hit zero, the LOWEST on record.
And in the spot market the fee has gone negative:
Meaning, some smelters now pay miners just to keep their furnaces fed.
By June, the spot fee sat at negative $127 a tonne.
This has arguably been one of the most consequential 20-day periods of America’s history.
Two interventions to save the Treasury market.
Both failed.
We may look back on this as America’s emerging-market moment.
None of us own enough hard assets.
https://t.co/RnZqdPy0sK