Recognized by NASA, I Reveal the Secret War of Artificial Intelligence

Last Friday, I joined Futuricast #258, one of Brazil's most-watched tech podcasts, for an hour-long conversation about the silent war between artificial intelligence companies — who funds it, who profits, and why it will cost far more than anyone imagines.

From theater to technology

Before writing a single line of code, I wanted to be an actor. I spent my early career in performing arts, until I understood Brazil doesn't offer the support to live off it. Technology came in as a practical alternative — but what kept me there was curiosity, not money.

The first thing anyone who wants to understand technology needs is curiosity. It's all very abstract — you can't see it, you can't touch it.

The AI war: why there's a billion-dollar race

Every AI company followed the same playbook: subsidize the product to break the adoption barrier. ChatGPT only went mainstream because it launched free. Nobody would pay the real cost of running an AI infrastructure — estimated at $1 trillion by 2030 — without dependency being built first.

It's the "here's a little treat" model: release everything for free, let the user get used to it, and normalize payment with every feature cut. Paying $700 or $1,000 a month for an AI sounds absurd today. It won't for long — because the professional who learned to work with AI can no longer deliver without it.

The numbers that sum up the conversation

• $1 trillion — projected global cost of AI infrastructure (cloud, data centers, chips) by 2030.
• $200 million+ — what some companies have already spent on AI before cutting staff to speed up the process, per accounts discussed on the show.
• $1 billion in 24 hours — revenue from China's biggest e-commerce livestream, a direct example of the scale AI makes possible.
• 40,000 iPhones — the number rented out by a single gadget-leasing company in Brazil, a sign of how big the market behind the tech "hype" really is.
• $12,000 (R$60,000) — price of an Nvidia local AI mini-computer, capable of running models without relying on external servers.
• 2 to 3 months — how long I spent testing before finding the vulnerability NASA recognized.
• 4 months — the gap between submitting the security report and NASA's official response.

Model or neural base? How an AI is actually built

"Developing an AI" doesn't mean building a brain from scratch — that would take years of research and massive teams. There's a neural base (the mathematical replica of how the brain forms synapses), often shared across companies. The real work is reprogramming that base for a specific behavior — which becomes a model. When a frontier model proves too powerful, it gets reined in: that's what happened with Mitos, whose public version, Fable, was already born with guardrails — a security decision that involved the U.S. government itself.

The vulnerability that became an official NASA letter

It started with no ambition: "Let me test my own skills." I picked large companies — the bigger, the more interesting the challenge. I spent two to three months testing, sleepless nights, using my own local AI as a copilot: I'd map out possible paths, it would test one in parallel while I explored another.

I found a flaw that allowed sending and receiving information that should have been protected. I never tried to access user data — the goal was to document, not exploit. I put together a full report: how I got in, what I found, how to fix it. I submitted it through NASA's official Vulnerability Disclosure Policy channel and moved on.

Four months later, an email. Formal recognition, hand-signed by a Senior Agency Information Security Officer. They confirmed the flaw was fixed and retested. It's not about the letter itself — it's validation that the knowledge I took from Brazil abroad carries weight where it matters most.

They could have never answered that email. It took four months, but they did. That's pride for Brazil — my foundation, my education, started here.

— Victor Fornitani, on the NASA recognition

I told this story in more detail — including the full official letter — in "NASA Sent Me a Letter. This Is What Real Security Looks Like."

Real risks: deepfakes, social engineering, and quantum computing

• Deepfakes and voice cloning: it's already possible to clone a voice and call a family member asking for money. The simplest defense? Create a "safe word" with your family — something only you know.
• Prompt injection: if an AI can't tell a legitimate command from a malicious one, anyone who replicates the right command can trigger the same action — including financial transfers.
• Social engineering is still the easiest door in: recent breaches affecting millions of accounts started with a credential obtained through social engineering, not a sophisticated code exploit.
• Quantum computing is a real threat to Bitcoin: old public keys, exposed by design, could be brute-forced in months — not trillions of years. All banking infrastructure built on SHA cryptography is, in theory, subject to that break.

"Mediocre plus" or "expert plus": AI amplifies who you already are

AI doesn't create competence — it amplifies whatever already exists. If you're excellent at what you do, it makes you an "expert plus." If you have no foundation, it makes you a "mediocre plus": faster, higher-volume, equally shallow. It needs prior human knowledge to function — every generative AI draws on human history to create anything new.

With great power comes great responsibility. It's not about having firepower — it's like handing you a bazooka. The question is: how are you going to use it?

— Victor Fornitani

Inside the CSX ecosystem: how the verticals complement each other

I use the analogy of a house: Nexforge builds the structure — the systems engineering that makes everything work. IronBit raises the wall, installs the cameras, makes sure cybersecurity protects that structure. DataZ understands how the house is used — data intelligence to reveal operational and commercial opportunities. And AUGE is our ecosystem for infoproducers: less a hosting platform, more a growth partner. See the full CSX Tech ecosystem — every vertical in detail.

Key takeaways from this conversation

• Free technology is never free — it's dependency being built before the billing starts.
• An AI model isn't a new brain, it's a reprogrammed one — understanding that difference changes how you evaluate any AI tool you try from now on.
• The most dangerous vulnerability is almost always human, not technical — trained behavior beats any firewall.
• AI amplifies whoever you already are — it doesn't replace competence, it multiplies it, up or down.
• Data, not apps, is this decade's real asset — whoever retains the most data controls the game, not whoever ships the prettiest app.

Technology is about solving problems, not ego

I've made this mistake before: building the most beautiful, technically sophisticated thing — for whom? For myself, for my ego, because I could pull it off that way. The question most people forget to ask is simple: does this solve a real problem? Frontier technology without purpose is just expensive vanity.

P.S. — Watch the full conversation on Futuricast #258. If you want to know more about AUGE or CSX Tech, the links are right above.