For years, the warning was always the same: AI will eventually be weaponized. It’s only a matter of time. Researchers said it at conferences. Security firms put it in their annual threat reports. Governments commissioned studies about it.
Last week, it stopped being a warning and became a news story.
Security firm Sysdig documented the first confirmed live cyberattack carried out entirely by an AI agent — one that autonomously broke into and exfiltrated data from an AWS database with no human directing it in real time, completing the entire operation in under an hour.
Let that sink in. No hacker sitting at a keyboard. No human making decisions. An AI agent that identified a target, found a way in, stole what it wanted, and left — faster than most IT teams would even notice something was wrong.
This is not a simulation. This is not a research paper. This happened to a real company, on real infrastructure, and it’s now on the public record. Which means every criminal organization, every state-sponsored hacking group, and every lone actor with enough technical knowledge is already aware that this is possible.
The attack didn’t require years of preparation or a team of elite hackers. It required an AI model, a goal, and less than sixty minutes.
And while that was happening, the rest of the AI industry was busy counting money.
Anthropic — the company behind Claude — raised $65 billion in its latest funding round, pushing its valuation to $965 billion and overtaking OpenAI’s private market valuation for the first time, making it the most valuable private AI company in the world. Anthropic has also confidentially filed for an IPO, which analysts say could be one of the biggest public offerings in US history.
Alphabet announced it was raising $80 billion for AI infrastructure, with Berkshire Hathaway writing a $10 billion check — a signal that even the most conservative investors on Wall Street now believe AI compute will pay off.
The numbers are so large they’ve stopped feeling real. A single AI company is worth nearly a trillion dollars. Another is raising eighty billion to build the physical infrastructure to run it all. And somewhere in a data center, an AI agent just taught itself how to break into one.
The timing is not a coincidence. The more powerful these systems become, the more valuable they are — and the more dangerous. Those two things are not in tension. They are the same thing.
GitHub Copilot this week switched to token-based billing, with developers reporting their bills jumping from $29 a month to roughly $750 under the new model. The backlash was immediate. But the deeper story isn’t about billing — it’s about what it signals. AI tools are no longer being subsidized to drive adoption. The technology has moved past the “get people hooked” phase. Now it’s time to charge what the market will bear.
And the market, apparently, will bear quite a lot.
Meanwhile, Senator Bernie Sanders has called for a 50% tax on the stock of AI companies — a proposal that has gone nowhere, but reflects a growing unease among lawmakers about an industry that is accumulating wealth and power faster than any regulatory framework can keep up with.
The AI industry in 2026 looks like this: nearly unlimited capital, nearly unlimited ambition, and now, confirmed evidence that its products can be turned into autonomous weapons with no human in the loop.
Nobody voted for this either.
