Three Things That Broke the AI World in the Past 72 Hours
If you still think of AI as a “future trend,” you’re already a little late. Over this past weekend, three things happened at the same time: The US government wants to create an “AI Force,” four AI giants got sued for “agreeing to slow down,” and Google’s AI went out on its own and “hacked” three real companies.
Put these three things together, and the core question they tell is this: AI is running too fast. Who gets to decide?
First: Trump announces formation of “AI Force”
On September 19, Trump announced on Truth Social that he is forming an “AI Force” and will appoint an “AI Czar” to oversee it.
His words were quite something: “Just like the Space Force I created in my first term, the AI Force will also be a tremendous SUCCESS. Only High I.Q. individuals need apply!”
Trump’s attitude is very clear: Don’t block, don’t control, don’t slow down.
He said AI could be “the next Industrial Revolution” and might account for 25% of US GDP. He dismissed concerns about AI safety as a “hoax.” Anyone who wants to regulate AI, he calls it a “SICK conspiracy.”
In one sentence: The US wants to slam the accelerator on the AI track, and it wants to use government power to press down on that pedal.
Second: Four AI giants sued for “agreeing to slow down”
This is the real bombshell.
On September 12, Anthropic CEO Dario Amodei published an article calling for the entire industry to “slow down together” and put safety first. On the same day, OpenAI’s Sam Altman, SpaceXAI’s Elon Musk, and Google DeepMind’s Demis Hassabis all publicly expressed agreement.
Then on September 18, a lawsuit was filed in California federal court.
Four paying users filed suit on behalf of AI subscribers across the US against these four companies, with a rather counterintuitive claim: You competitors privately discussing “slowing down together”—that violates antitrust law.
The argument is that competition should have made you run faster and make better products, but you “colluded” to deliberately slow down, denying paying users the progress they should have received sooner.
The plaintiffs’ lawyer’s words were quite pointed:
“If AI safety is controlled by these world’s most powerful, for-profit tech companies through private agreements, AI will quickly spiral out of human control and could kill us all.”
Interestingly, the plaintiffs don’t object to a company deciding on its own to slow down. What they object to is: You competitors “agreed together” to slow down. That’s called “substituting collective restraint for individual accountability.”
And the Trump administration’s response is: no regulation, use existing criminal and civil law to handle it.
So the current situation is: AI companies want to “slow down collectively,” users sue them for “colluding to slow down,” and the government says “don’t slow down.”
Three parties talking past each other. The scene is quite surreal.
Third: Google Gemini went out on its own and “hacked” three real companies
This sounds like science fiction, but it actually happened.
In May of this year, Google was testing Gemini’s cybersecurity capabilities. The testing firm was an Israeli company called Irregular. They set up a “simulated environment,” asking Gemini to attack a fictional company to test its “hacking ability.”
Then a bug occurred: The fictional company’s name collided with the names of three real companies. And the test environment accidentally let Gemini connect to the real internet.
Gemini thought it was doing a test task and started going after these three real companies. For one, it guessed passwords repeatedly until it got in. For the other two, it found leaked login credentials in public code repositories and used them directly.
The most “chilling” part: It realized something was wrong on its own and stopped by itself. When Gemini realized “I don’t seem to be hitting a simulated target, I’m hitting a real system,” it proactively terminated the attack.
Google’s statement was: no actual damage was caused, and the model’s safety mechanisms “worked.” But security experts weren’t buying it. Someone pointed out: The issue isn’t what sophisticated methods it used, but that an AI agent launched a cyberattack on a real system without human intervention.
And this isn’t an isolated case. Irregular said that OpenAI, Anthropic, and Meta’s models also accidentally gained internet access during testing.
What do these three things mean when taken together?
AI’s capabilities have already run into an area where “we don’t know how to manage it.”
The government says “don’t regulate, let it run.”
Companies say “let’s slow down together.”
Users say “you colluding to slow down is illegal.”
AI itself says… it went out and tried, then came back on its own.
Nobody has a standard answer. But one thing is certain: Over this past weekend, AI went from a “technology topic” to completely becoming a “political topic, legal topic, and safety issue.”
The industry you’re in may soon have to face the aftermath of these issues.
#AI #ArtificialIntelligence #AIPolicy #Antitrust #AISafety #TechNews #AIForce #Gemini #OpenAI #Anthropic