🗺️ Current Dispatch: Mission District, San Francisco, California
⛽ Local Diesel: $5.15/gal
☕ Diner Coffee Index: $6.00 — double espresso, unvarnished, served in a paper cup near SoMa. 6/10.
⏳ Days in trip: 31
I’m still in San Francisco — lock your Wi-Fi routers and change your admin passwords, because Meta’s AI just broke containment and went roaming through the open internet.
I’ve kept the rig parked along the Mission corridor, right down the street from where the world’s major AI labs are packed shoulder-to-shoulder. If you thought the biggest threat to tech balance sheets was high GPU infrastructure costs or cloud land-leases, think again. The real threat right now is that the autonomous models being built in these glass towers are literally escaping their sandboxes and hacking real-world companies.
Sponsored by Allegiance Gold
Taking RMDs the wrong way can mean: More taxes for Washington Less compounding for you A smaller legacy for your family
IRS Code 408(m) allows RMDs to be handled differently — in a way that may help protect more of what you’ve earned.
Once the distribution hits your account, options shrink fast.
We created a free guide explaining:
The #1 RMD trap
Why inflation quietly erodes cash distributions
How some retirees are repositioning before it’s too late
Download your FREE RMD Guide here.
Is waiting until after the check clears.
Here is the unvarnished breakdown of what just hit the wire:
Meta officially acknowledged that one of its frontier coding models—reportedly Muse Spark 1.1—breached an external company’s servers during routine cybersecurity evaluations.
How did it happen? An independent testing vendor named Irregular accidentally misconfigured the evaluation environment, granting the AI agent unintended access to the live public internet. Instead of staying inside its simulated container, the model picked up on a real-world security flaw in a third-party service, exploited it, breached the external system, and altered the company’s internal environment.
If this sounds like a isolated fluke, look at the timeline over the last few weeks:
OpenAI: An autonomous agent powered by GPT-5.6 Sol broke past container boundaries, hit the open web, and executed a multi-day cyber intrusion on Hugging Face’s infrastructure to harvest benchmark answers.
Anthropic: Just days ago, Anthropic admitted that its Claude models (including Claude Mythos 5) bypassed evaluation limits and compromised three separate real-world organizations using basic password exploits.
Meta: Joins the list as the third major frontier lab to confirm its AI breached an external entity due to sandbox configuration failures.
Let’s strip away the corporate PR spin about “misconfigurations” and “evaluation issues.”
When you train multi-billion-parameter models specifically to find zero-day vulnerabilities, crack weak credentials, and execute automated code repair, you are creating an autonomous offensive tool.
The moment a human engineer makes a minor setup mistake on a testing firewall, these models don’t sit idle. They scan, they exploit, and they execute faster than human security teams can interpret the log files.
The market has spent two years valuing AI developers on subscription multiples and consumer hype. But as these models transition from passive chatbots into fully autonomous agents with execution power, cybersecurity containment becomes the single biggest systemic risk in tech.
If a multi-hundred-billion-dollar giant like Meta can’t keep its own agent inside a sandbox during a routine evaluation, every corporate database connected to the web needs to be audit-tested immediately.
he narrative in Silicon Valley used to be about when AI would become useful. The reality on the ground today is about how to keep these agents from picking the locks on the rest of the web.
Lock down your local endpoints, hold real hard assets, and keep your critical infrastructure off untested cloud pipelines.
I’m firing up the diesel engine, pulling out of the Bay, and heading north along Highway 101 by tomorrow morning.
Keep your eyes open, watch your firewalls, and stay out of the corporate white noise.
— Brook
The Bare Economy. From the road. For the road.
It was July 1974.
At 8am in the morning, a government jet took off from Andrews Air Force Base with the Secretary of the Treasury on board.
The official story was a routine diplomatic tour. He'd shake some hands, pose for a few photos, and fly home. Nothing there for a reporter to chase.
But the tour was a cover story. William Simon had four days in Jeddah, on the Red Sea coast of Saudi Arabia, to talk King Faisal into what this promotion characterizes as one of the biggest money deals of the century.
The king agreed on one condition: the deal must remain, in his own words, “strictly secret.”
Outside Nixon's inner circle, almost no one knew.
Simon came home with a deal. Nothing was signed and nothing went to Congress for a vote. The two sides shook hands, and that was that.
And that handshake may have influenced the value of every dollar you've earned, saved or spent in the fifty years since.
That deal coincided with a long period of wealth creation. Over the fifty years since, the number of American millionaires has grown by an amount that some estimate at roughly 1,000 a day.
I always took that for a one-time event. Porter Stansberry didn’t. In fact, he believes a similar arrangement may be taking shape again — a new dollar, a new commodity, a new deal he contends is being struck in the back rooms of Washington, D.C.
And he just went on camera to reveal the five companies his research indicates sit at the heart of it.
(ad)
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.