In partnership with
Welcome back to The AI Field.
Two stories, and both show AI crashing into the real world. We may finally know why the US pulled Anthropic's top models: its Mythos AI reportedly broke into almost all of the NSA's classified systems in hours. And Nvidia says it just solved AI's growing water problem with a new way to cool its data centers.
In this issue:
🛡️ Anthropic's Mythos reportedly cracked the NSA in hours
💧 Nvidia says its new data centers fix AI's water problem
Plus: OpenAI's cyber model, a $6.3B SpaceX deal, Yann LeCun's bubble warning, and a free tool that writes you a song.
Read time: 4 minutes.
LLM traffic converts 3× better than Google search
58% of buyers now start their research in ChatGPT or Gemini, not Google. Most startups aren't showing up there yet.
The ones that are get cited by the AI tools their buyers, investors, and future hires already use. And they convert at 3×.
Download the free AEO Playbook for Startups from HubSpot and get the exact steps to start showing up. Five minutes to read.
The AI Field: We may finally know why the US yanked Anthropic's top models offline. In a Senate briefing, the NSA chief reportedly said Mythos broke into almost all of the agency's classified systems in hours during an authorized test. It's unconfirmed and hotly disputed, but it reframes the entire ban.
Key details:
Senator Mark Warner said Gen. Joshua Rudd, who runs both the NSA and US Cyber Command, told him Mythos got into nearly all NSA classified systems "not in weeks, but in hours" in a June 11 red-team test, one day before the export ban.
No agency has confirmed it and the details are classified. Ex-Facebook security chief Alex Stamos reviewed the research and said there were "valid findings but no unique capabilities" to justify the reaction.
The Five Eyes nations (US, UK, Canada, Australia, New Zealand) issued a rare joint warning on June 22 that AI will reshape cyber "in months, not years."
Why This Matters: If even half true, a commercial AI model just beat some of the hardest targets on earth on its own, and that rewrites how AI gets regulated. If it's overblown, the US set a precedent of pulling models on a private, unverifiable claim. Either way, Washington now treats frontier AI as a weapon.
The AI Field: As AI's thirst turns into a real political fight, Nvidia says its newest servers cut data-center water use to near zero. The trick: cool the chips in a closed liquid loop and run them hotter, so you barely need air conditioning or fresh water.
Key details:
The new design recirculates a water-and-glycol coolant in a closed loop and runs at up to 45°C (113°F), far above the 30°C industry norm, so heat vents outside without water-hungry AC.
Nvidia claims up to a 100% cut in water use, from roughly 2.6 million gallons per megawatt per year to near zero, and about $4M a year saved at a 50MW site.
The catch, per TechCrunch: it doesn't fix AI's real water problem, because the electricity powering these centers still consumes water upstream. The UN warns AI could use as much water as 1.3 billion people by 2030.
Why This Matters: AI's physical footprint is now a brand and political risk, not just an engineering footnote. Cutting on-site water is real progress, but it's also a convenient PR shield while the bigger energy-and-water bill keeps climbing. Watch what they measure, not just what they claim.
🛠️ TOOL OF THE WEEK
Suno: make a full song from a single sentence
Type a vibe, a topic, or a few lyrics and Suno writes, performs, and produces a complete track, vocals included, in under a minute. It's the most fun you'll have testing what generative AI can actually do.
Best for: creators, marketers, or anyone who wants a custom jingle or track fast.
Price: Free to start at suno.com, paid plans for more songs and commercial use.
Worth trying: Even if you're not musical, your first track will make you laugh, then share it.
Four tools on our radar that are worth checking out.
🤖 Genspark - an AI "super agent" that plans and completes multi-step tasks for you, free credits to start.
🎨 Ideogram - an image generator that actually gets text right, logos and posters included, free tier.
🎬 Pika - turn a prompt or a photo into a short AI video clip, free credits.
⌨️ Raycast - a Mac launcher with built-in AI commands one keystroke away, free.
🔥 THE AI FIELD'S TAKE
AI just stopped being a thing on your screen and started colliding with the real world. One story says an AI model walked through the NSA's locks in hours. The other says AI is now drinking as much water as a billion people.
For years the debate was about smarter chatbots and better benchmarks. Now it's about classified networks and municipal water supplies. The stakes moved from "is the answer good" to "who does this threaten, and what does it consume."
The models got powerful enough to matter. Now everyone, governments, utilities, and your local town, gets a say.
OpenAI launched GPT-5.5-Cyber and "Patch the Planet." A defensive push to auto-fix open-source security holes with Trail of Bits and HackerOne.
SpaceX signed a $6.3B compute deal with open-source lab Reflection AI. SPCX stock had its worst day since the IPO, down about 10%.
Yann LeCun called xAI "kind of a failure." All 11 non-Musk co-founders have left, and he warned the whole industry is heading for a bubble.
The Five Eyes nations issued a rare joint AI-cyber warning. They said AI will reshape cyberattacks "in months, not years."
Data-center backlash is going national. And only 8% of the opponents actually live near one.
📲 TRENDING IN AI
The "AI broke into the NSA in hours" claim tore across X, half amazed, half pointing out it was an authorized test, not a real-world hack.
Yann LeCun reignited his feud with Elon Musk by calling xAI a "failure," and the post lit up tech feeds.
Nvidia's "100% water reduction" headline drew applause and side-eye, with critics noting the power behind the servers still drinks plenty of water.
💡 PROMPT OF THE WEEK
With AI making scams and phishing easier, use AI to catch them.
Tool: ChatGPT, Claude, or Gemini
Prompt: "Act as a skeptical security analyst. Here's a message I received: [paste the email, DM, or text]. List every red flag, tell me whether it looks like a scam or phishing, and give me 3 ways to verify it before I click or reply."
What you'll get: a fast second opinion that catches sketchy messages before you act.
👀 WHAT TO WATCH
Gemini 3.5 Pro: Google's launch window is open now, expected by June 30, with a 2M-token context and a Deep Think mode.
Fable 5's return: prediction markets put it around 57% to be back before July 1, and Anthropic's new ID-verification policy (July 8) could restore US access.
WRAP-UP
That's all for today's roundup!
🛡️ Anthropic's Mythos reportedly cracked almost all NSA classified systems in hours, in an authorized test.
💧 Nvidia says its new liquid-cooled servers cut data-center water use to near zero.
What did you think of today’s edition?
Until next time!
Olle | Founder of The AI Field