Friday, October 2, 2026 · 5 min read

OpenAI has fired three of its own safety researchers for passing information to an outside safety group. One congressman's first reaction was to call them whistleblowers.
Broadcom is lending Anthropic up to $42 billion to rent Broadcom's chips. Amazon wants to sell its Nvidia chips to investors and rent them back.
Let Meerkat explain.
OpenAI fires three safety researchers, and a congressman cries whistleblower
Broadcom and Amazon turn chip sales into loans and leases
The 10-year yield's 24-year high, in one chart
The jobs report this morning, and Anthropic's IPO calendar
SCOREBOARD
NVIDIA | MICROSOFT | ALPHABET |
META | AMAZON | ORACLE |
S&P 500 | NASDAQ | VIX |
Thursday close. S&P up 0.2%, with Nvidia up 1.1% and Alphabet down 1.7%. On the week: Nvidia +2.8%, S&P −0.5%, Nasdaq −0.3%. Prices via CNBC.
ON WATCH
Three safety researchers, shown out

OpenAI has fired three safety researchers it accuses of passing confidential company information to an outside AI safety group. The Wall Street Journal broke the story on Thursday, and OpenAI hasn't named the group or said what was shared.
It lands in the same week OpenAI said it had warned more than 100 organisations that its testing agents may have bypassed their security or otherwise hit them. It's still reviewing about 50 petabytes of records to find the full list.
Congressman Greg Casar, who chairs the Congressional Progressive Caucus, wasted no time. "Looks like they're firing whistleblowers. What are they hiding?" he posted, promising to send OpenAI a demand for transparency.
OpenAI says its investigation confirmed the three "mishandled sensitive information outside established company procedures." It says that broke "the trust essential to our work."
Set against that, OpenAI has given staff a formal way to raise alarms: it set up a misalignment reporting process on 16 Sep. It's also the one publishing its agents' incidents, which most companies would rather keep in a drawer.
MEERKAT'S TAKE: A company can sack people for leaking and still be in the right. But these were the safety team, and they took it to a safety group, so it's fair to ask what they saw that the official channel couldn't carry.
SPONSORED
Elon's new company is private. These 3 tickers aren't.
The next Apple may already exist. Insider sources say Elon has spent two years building a secret device inside Tesla's facilities — one he claims will be "10x bigger than the largest product in history."
There's just one problem: the company is private, and unless you know Elon personally, you can't buy a single share. That was true until Guardian's research team found three public ticker symbols sitting in the launch supply chain.
Click here to see all 3 tickers, free of charge.
You won't hear these names on CNBC — Wall Street hasn't published a word on the connection. But when the launch hits September 21, that quiet ends.
Some are already calling this the biggest opportunity since AI. For anyone who missed Apple before the iPhone, this may be a second look at that kind of setup.
THE MACHINE
The chipmakers become the bank

Anthropic's IPO filing shows Broadcom agreeing to lend it up to $42 billion to lease Broadcom's own chips, Reuters reported. That could cover about a third of a $125.2 billion, five-year deal for Google's TPU chips, and Broadcom can swap the debt for Anthropic shares.
The same day, the Financial Times reported that Amazon wants to put about $8 billion of Nvidia chips into a separate company. Investors would buy in through debt and up to 10% of the equity, and Amazon would rent the chips back.
Let Meerkat translate. When the seller lends you the money to buy from the seller, the sale looks bigger than the cash behind it. Anthropic's own filing flags "potential conflicts of interest" in Broadcom being both supplier and lender.
The bull case is that the demand behind it is real: Broadcom expects about $115 billion of AI chip sales in fiscal 2027 and $230 billion the year after. Anthropic has also put cash in a locked account for Broadcom as security.
QUICK PAUSE. WHY SHOULD YOU CARE? Normally a chipmaker's risk ends when the customer pays. With loans like these, an AI lab's bad year becomes the chipmaker's bad year too.
SOMEONE SMART SAID

The AI build out is potentially creating inflationary pressures that may not resolve very quickly.
Cook said it on Thursday at a New York Fed event, calling AI one of the main things that concerns her for 2027. Companies have announced about $2 trillion of AI investment plans, mostly data centres, and she pointed to costs like electricity, with key inputs each up around 5% in a year.
Cook voted for last month's rate hike, and the Fed fights inflation with higher interest rates. If AI building keeps prices up, that's an argument for keeping rates high.
She still expects AI to lift productivity in the long run. Her worry is timing: "when the productivity gains that would produce disinflation will come."
THE CHART
Borrowing costs hit a 24-year high

US Treasury daily par yield curve, 10-year, month-end; latest close 1 Oct 2026
The 10-year Treasury yield, the rate that steers mortgages and company loans, touched 5.34% on Thursday, its highest since 2002. It closed at 5.24%, according to the Treasury's own figures.
That's the rate the AI building boom borrows against. Data centres, chip leases and deals like Broadcom's are all priced off it, so every borrowed dollar costs more.
In fairness, the market took it in its stride. Yields eased off the high and the S&P 500 still closed up 0.2%, with Nvidia up 1.1%.
WHAT TO WATCH: If the 10-year closes above 5.34%, that's a fresh 24-year high and borrowing for data centres gets dearer again. If this morning's jobs report comes in soft, yields have room to ease.
HEADS UP
California subpoenas OpenAI · Attorney General Rob Bonta wants more answers on OpenAI's cyber incidents, saying AI firms have "a moral and legal responsibility" not to enable cyberattacks.
Google wins its AI Overviews fight with publishers · Judge Amit Mehta dismissed antitrust suits from Penske Media and Chegg, ruling that "an expectation is not an agreement."
Anthropic sets an IPO calendar · It's meeting big investors on 14 Oct and could list as soon as mid-November, at a valuation of up to $2 trillion, Bloomberg reports.
$300 million of AI servers allegedly smuggled to China · Federal agents arrested a California company owner accused of routing export-controlled servers through Malaysia and Singapore in 2023 and 2024.
OpenAI and Synopsys team up on chip design · They'll build GPT-Synopsys, a model meant to run chip-design software like an experienced engineer. Early tests with chipmakers are underway.
ON THE HORIZON
When (ET) | What |
|---|---|
Fri 2 Oct · 8:30am | September jobs report. The first big data point since the 10-year hit its 24-year high |
Tue 6 Oct | Anthropic appears before Australia's parliamentary inquiry into AI, in Canberra |
Wed 14 Oct | Anthropic's pre-IPO investor day at its San Francisco headquarters (Bloomberg) |
Wed 28 Oct · 2pm | Fed rate decision. Dallas Fed chief Lorie Logan says rates may need to rise 50 basis points or more |
Mon 9 Nov | Week Anthropic could start marketing its IPO, aiming to list before Thanksgiving |
Fed dates via the Fed.
MEERKAT ON YOUTUBE
WTF Is Happening To The AI Bubble? Everyone's watching the stock market for the crack, but the first one shows up somewhere almost nobody looks: the price of insuring a company's debt. For one of the biggest names in AI, that price just hit an 18-year high. Watch on Meerkat Explains.
THE BURROW
Quick vote. OpenAI fired three safety researchers for sharing information with an outside safety group. What's your read?
OpenAI fired three safety researchers for sharing information with an outside safety group. What's your read?
Trivia. In May 2024, OpenAI's "superalignment" team, set up to keep future AI under control, lost both its leaders. One was co-founder Ilya Sutskever. Who was the other? Answer's at the bottom of the email. No peeking.
ALSO WORTH YOUR INBOX
The daily crypto brief from the same burrow. Crypto Nutshell is Bitcoin, the news and the numbers in a five-minute read.
HOW DID WE DO?
One tap, and Meerkat listens. How did today's issue land? Hit a button, and add a line if you want to tell Meerkat why.
How did Meerkat do today?
Back Monday at 6:30. The robots work weekends, so you don't have to.
Somebody in your life works in compliance? Forward them this.
Meerkat is not a financial adviser, and nothing here is advice. Meerkat is a small animal with a whiteboard and a confidentiality agreement it hasn't signed. Please do not sue Meerkat.
TRIVIA ANSWER
Jan Leike. He left saying safety had "taken a backseat to shiny products" and joined Anthropic weeks later.



