Support Us & Upgrade to the Operator Edition

Today, a single hacker turned cheap AI agents into a break-in machine and walked through the digital front doors of John Deere, SharkNinja, and Cummins. Meanwhile, prosecutors say the ex-OpenAI engineer caught bugging an executive boardroom had a Saudi royal in his DMs. Two very different stories, same lesson: AI’s biggest security hole right now is people, and the tools they’re pointing at you.

Your AI is only as good as what you feed it. Feed it a transcript with your product names, acronyms, and numbers spelled wrong, and you spend the afternoon hand-fixing the output.

Wispr Flow Notetaker uses your dictionary and calendar, so product names, acronyms, numbers, and uncommon names come out spelled right. It pops up when your meeting starts and captures Zoom, Google Meet, Teams, and Slack huddles with one click, no bot joining the call.

Then your meetings show up inside Claude or ChatGPT through the built-in connector. No copy paste. Try Notetaker free on Mac and Windows.

Try Notetaker Free

Google just released Gemini 4 Argon, and in the launch episode of Release Notes, the team revealed how they actually use it day to day. One workflow stood out: they don’t let it write for them. They use it as a ruthless copy editor. Here’s how to steal it:

💡 Pro tip: If a teammate sends you something that feels too polished, borrow Google’s go-to reply: “Share your prompt with me.” Refining the prompt together beats accepting a one-shot output.

Google’s internal cloud sales team is one of Argon’s heaviest users. This isn’t just a coding model.

Watch the full Gemini 4 Argon episode here

Want the version of this that actually runs in your business?

The 30-Second Play gives you the move. The Operator Brief gives you the full system: prompt libraries, API cost math, and the workflows our subscribers are using to replace entire freelance line items.

This week: How to save money on lawyers and document SOP’s with AI

👉 Upgrade to The Operator Brief

🚀 Elon Musk says he’ll give one SpaceX share to every kid in America next year, as “universal basic capital” quietly becomes the idea both parties are rallying behind to soften AI’s hit on jobs. The real question is whether a single SpaceX share solves or even mitigates structural unemployment.

🤝 Likewise, Anthropic is pledging 5% of its shares to a new public benefit fund aimed at workers displaced by AI, just weeks before a potential blockbuster IPO. Anthropic is acknowledging the harm its own product could cause before it cashes out. Once again, whether 5% is enough is a separate conversation.

💻 Amazon unveiled TorchNeuron, a free tool that lets PyTorch code run on its Trainium chips with one config change, taking direct aim at the CUDA moat behind Nvidia’s dominance. Amazon is handing developers an exit ramp from Nvidia’s ecosystem and they made it as frictionless as possible.

💰 Jim Keller’s Tenstorrent raised $1.5 billion and more than doubled its valuation to $6.2 billion, and he’s designed chips at Apple, AMD, Tesla, and Intel. When the same person keeps showing up at every inflection point in chip history, it’s worth paying attention. And it’s another sign investors are betting big on open alternatives to Nvidia.

🇨🇳 DHS is ordering every federal civilian agency to remove DeepSeek and anything built on it, citing the risk that government data ends up in Beijing’s hands. The US government is drawing a hard line between American AI infrastructure and Chinese models. This ban won’t be the last one.

The ex-OpenAI engineer accused of bugging an executive conference room appears to have had considerable support from the Saudi government, according to a court filing made public Monday.

Prosecutors laid out the details while arguing that Abdulrahman Alfozan, 35, should stay in custody as a flight risk. Here’s what the filing alleges:

He currently faces one federal charge for intercepting oral communications, with a maximum of five years. Prosecutors say more charges could follow. He has not been convicted of anything.

This isn’t new territory. In 2019, two Twitter employees were charged with spying for Saudi Arabia by pulling user data on government critics. And on the very same day as this filing, the State Department sanctioned five Chinese entities tied to an ex-Google engineer accused of stealing Google’s AI infrastructure secrets.

Here’s the asterisk: everyone’s been guarding the model weights. But look at what was allegedly targeted here: not code, but conversation. The room where executives discussed strategy, technology, and government engagements. In 2026, what AI leaders say behind closed doors may be worth more than any file on a server.

🔮 Prediction: Frontier AI labs are about to start looking a lot more like defense contractors.

The logic is simple. Once a company’s internal discussions become a matter of national interest, the security model has to match. Expect compartmentalized access to sensitive meetings, physically secured rooms, stricter background checks for certain roles, and much tighter controls on internal tools that reveal who meets where and when. The irony that OpenAI’s own productivity tool was allegedly used for reconnaissance won’t be lost on its security team.

Washington will likely accelerate this. With labs this central to policy and defense work, I’d expect security requirements to start showing up as conditions for government contracts.

The hard part is the trade-off. These labs run on global talent, and the vast majority of foreign-born researchers are exactly who they say they are. Overcorrect, and you push brilliant people toward competitors abroad, which is precisely what that “Saudi needs you” text was trying to do. The labs that solve security without turning into a fortress will keep their edge.

💬 Should AI labs vet employees like defense contractors? Or does that kill the open culture that made them great? Reply and pick a side.

Read the full report here

A single hacker used autonomous AI agents to break into more than 100 company websites, and it cost about as much as a used car.

Security firm NorForge published its investigation Monday after finding more than 20 exposed control panels for Cairn, an AI hacking system you give a target and a goal, and it coordinates agents to go find a way in.

Here’s what the panels revealed:

The victims are responding very carefully. Deere says it has no indication sensitive data was accessed, SharkNinja says an attack on a product support site was contained, and Cummins is investigating a possible breach through a third-party vendor.

The targeting was almost industrial. One automated task literally read: confirmed real shopping sites with 200,000 to 1.5 million in traffic, already tested to make sure orders could be placed, now go find a way to plant a webshell. NorForge checked one claimed intrusion on a hardware parts vendor. The webshell was live.

Here’s the uncomfortable part: the hackers didn’t lock their own doors. These panels were wide open on the internet, exposing everything the agents found, including hundreds of victims’ passwords and API keys (11% verified as still valid). Anyone who stumbled onto one could, in theory, redirect the agents to new targets. Even legitimate red teams were caught running real client tests on unsecured panels. And my favorite detail? A buggy script kept chopping the “w” off domains, turning “wholesale” into “holesale,” which is how NorForge linked the panels together.

🔮 Prediction: The real story is not that AI can hack. We knew that. It’s the price tag.

For decades, hacking a mid-sized e-commerce store wasn’t worth a skilled human’s week. At $50 a breach, the math flips. Every store doing a few hundred thousand visits a month just became an economically rational target, and that’s precisely the band this group went after. The long tail of the internet, the businesses that assumed they were “too small to bother with,” is now the most attractive real estate.

Two things follow. First, patch speed becomes a survival metric. When an agent can grind on your site for 23 hours straight without coffee breaks, a known vulnerability left open over a long weekend is a door left open. Expect cyber insurers to start asking about this explicitly, and pricing it.

Second, watch the policy whiplash. On Friday, NIST, the NSA, and CISA issued 10 practices for containing rogue AI inside frontier labs. Useful, but this attack didn’t need a frontier lab. It needed a cheap open model, an off-the-shelf harness, and a credit card. And the very same day this broke, DHS ordered DeepSeek ripped out of federal civilian systems. I’d bet this case gets cited in every open-weights debate in Washington for the next six months.

The defenders who win will be the ones who run these agents against themselves first.

💬 Has your company pointed an AI agent at its own website yet? Hit reply and tell me. I’m genuinely curious how many of you have.

Read the full NorForge investigation here

Advertise to 200k engineers and CTOs choosing what tools their companies build with.

👉️ Learn how to sponsor this newsletter

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI