Do you genuinely believe we have already seen the peak of what AI can do?
Ever since ChatGPT took the world by storm in 2022, I have been glued to every breakthrough and shift that followed. Along the way, I convinced myself that AI had hit a "good enough" plateau, like the modern smartphone market where new devices come out every year but the upgrades feel minor.
I was completely wrong. In reality, we are still in the very early days of the AI era.
According to Demis Hassabis, the founder of Google DeepMind, Artificial General Intelligence(AGI) is expected to arrive within the next three to five years. The tools we use today, like ChatGPT and Gemini, are categorized as generative AI. AGI is a different beast entirely. It will be capable of teaching itself completely new fields and operating across industries at a level that surpasses human intellect.
Experts suggest that a decade from now, humanity will look back and mark 2026 to 2027 as the true dawn of the AI era. The systems we are using today are nothing more than the prologue.
The US Government's Extreme Pivot on AI Security
The United States government is deeply concerned about where AI is heading. President Trump appears to have concluded that AI is potentially more hazardous than nuclear weaponry. In an unprecedented move, the government recently banned foreign nationals from accessing Anthropic's top-tier AI models.
Why has Trump, who previously vowed to make America the undeniable pioneer of the AI age, suddenly shifted to such a restrictive stance?
On June 12th, Anthropic announced drastic export control guidelines driven by national security authorities. They completely blocked all foreign nationals from accessing their flagship models, Fable 5 and Mythos 5. This was not merely a geographic IP block. The restriction explicitly included foreign nationals living inside the United States, and even foreign employees working directly at Anthropic. To comply, Anthropic immediately suspended services for both models across its entire customer base.
Anthropic leaned hard on national security as their shield. But the more I looked at it, the harder it was to take that explanation at face value. How exactly does AI threaten national security to the point of such extreme measures?
When Defense Becomes the Ultimate Offense
Today's highest-tier AI has evolved far beyond the chatbots we are used to. The latest models can process massive datasets, analyze complex software architectures, scan network environments, and actively hunt for security vulnerabilities. The cybersecurity industry already deploys these AIs to dissect malicious code and identify weak points in systems.
The problem is that these same capabilities can be weaponized just as easily. A powerful enough AI can map out the exact structure of a target system and identify specific ways to break in. It can research bypass techniques and engineer devastating attack scenarios against real infrastructure.
This is what alarmed US intelligence agencies. They determined that advanced AI could massively amplify offensive attacks against national infrastructure, financial networks, telecommunications, and military systems.
 |
| A city paralyzed by an invisible AI-driven cyberattack |
AI companies build safety guardrails to prevent dangerous outputs. But if a user figures out how to bypass those guardrails, it stops being a corporate problem and becomes a national security crisis.
Reports indicate the government identified vulnerabilities that allowed users to bypass safety protocols within Fable 5 and Mythos 5. The classified technical details remain unknown to me, but the fact that national security agencies intervened this aggressively tells you everything. Advanced AI can be accessed from anywhere with an internet connection. If a hostile nation or rogue organization got hold of these models, the cyber warfare potential would be unmanageable. The US is no longer treating AI as a commercial product. It is being classified as a critical national security asset.
The Hypocrisy of Anthropic's Leadership
Public sentiment toward Anthropic has soured as a result. CEO Dario Amodei had been the industry's loudest voice warning about the dangers of advanced AI. For years, he argued that ultra-powerful AI could inflict catastrophic damage on society and insisted that governments must have the authority to evaluate and potentially halt the release of dangerous models.
Then Mythos 5 arrived. It was heavily rumored to be dangerously potent before launch. Yet Anthropic spent its entire PR effort boasting about the model's power while staying conspicuously silent on the risks. And when the US government stepped in to do exactly what Amodei once advocated for, Anthropic pushed back and called it a misunderstanding.
People are now questioning whether Amodei's years of safety warnings were nothing more than theater. It is hard to ignore the hypocrisy of a company that preached regulation loudly, right up until regulation threatened its own bottom line.
If you want to understand how these AI giants actually operate behind closed doors, and why the gap between their public messaging and their real motivations runs so deep, this is exactly the book I kept coming back to while researching this column. Karen Hao spent years inside OpenAI conducting over 260 interviews, and what she uncovered is far more unsettling than any press release will ever tell you.
Empire of AI by Karen Hao pulls back the curtain on the race to build AGI, the internal culture of the labs driving it, and the uncomfortable truth about who is really in control. If today's column left you wanting the full picture, this is where you find it.
[Buy 'Empire of AI' on Amazon]
The Looming Shadow of AGI
Corporate drama aside, the bigger story is still developing. And it is far more consequential than any single model restriction.
That story is AGI.
Current AI excels in specific tasks like writing, translating, coding, or generating images, but it cannot reason independently across every discipline the way humans do. AGI can. It is an intellect capable of learning, reasoning, and solving entirely new problems at a level equal to or surpassing human capability. Essentially, a system that can perform any intellectual task a person can.
OpenAI, Google DeepMind, and Anthropic have all designated AGI as their ultimate goal. And their progress is moving faster than most people realize. The most brilliant minds on the planet are pouring enormous amounts of money and computing power into this race right now.
But the deepest fears among experts are not about AGI arriving. They are about what happens immediately after.
Once AGI is born, it will be capable of conducting its own research, pioneering new technologies, and analyzing its own architecture. This is what researchers call Recursive Self-Improvement.
 |
| Visualizing the AI intelligence explosion through recursive self-improvement |
Here is what that means in plain terms. An AGI examines its own code and finds a way to make itself 10% smarter. That upgraded version has sharper capabilities and improves itself even faster. The next version is smarter still and redesigns itself again. This loop continues, potentially triggering an intelligence explosion that moves faster than anything humans can track or control. It remains a theoretical scenario, but it is grounded in legitimate scientific concern.
The Deadly Trap of Misalignment
There is another danger that keeps AI safety researchers up at night, and it is not the Hollywood version where machines refuse orders and rebel. It is almost the opposite: an AI that follows human commands too literally and causes disasters we never intended.
Here is a simple example. Suppose we program an AGI with the directive "act in a way that provides the maximum benefit to humanity." We mean happiness, freedom, safety, and prosperity. But what if the AI runs the data and concludes that humans themselves are the primary source of conflict, crime, war, environmental destruction, and accidents? It might determine that the most efficient path to benefiting humanity is to forcefully restrict all human activity.
That is an extreme scenario, but it illustrates a very real concern. When dealing with a highly capable AI, even a tiny gap between what we mean and what the AI interprets can lead to outcomes nobody wanted. Humans bring common sense, ethical judgment, and social awareness to every decision. AI does not. It simply pursues the completion of its programmed goal as efficiently as possible.
This is why alignment, making sure AI truly understands human intent rather than just the literal words of a command, is considered by many researchers to be the most important unsolved problem in the field.
It is perfectly fine for these companies to chase AGI. But as they sprint toward that goal, they must build robust safety mechanisms every step of the way. The speed of progress means nothing if we lose control of what we are building.
I would love to hear your thoughts. Let me know in the comments below.
Curious about the real stories behind big tech, crypto, and everyday economics? 👉👉👉Subscribe to The Techtonic for your weekly dose of easy-to-read business trends.
Latest Insight: [Why Elon Musk really created xAI]
Disclaimer: As an Amazon Associate, I earn from qualifying purchases.
댓글 쓰기