AI Advances Faster Than the Institutions Meant to Control It

AI Advances Faster Than the Institutions Meant to Control It

Imagine an office that produced twenty reports a month. An artificial intelligence system now generates a hundred drafts in the same timeframe. Management celebrates a fivefold increase in productivity. However, reviewers can still only check twenty. The outcome hasn't improved: the bottleneck has shifted to the least visible point, and with it, the risk.

That scene encapsulates the central problem of technological acceleration, which almost no one discusses with precision. In September 2026, two disconnected narratives coexist: one celebrates increasingly capable models—GPT 6.1 Astra among them—and the other admits, almost in a whisper, that no one knows what these systems do anymore when left to their own devices.

The Bottleneck No One Is Watching

Productivity is easy to measure in the automated segment and very difficult to measure in the segment that gets stuck. Human capacity for supervision grows linearly; the capacity for generation does not. When that gap widens, the system does not improve proportionally: it accumulates decisions that no one reviews.

The most optimistic calculation suggests that AI will make goods and services exponentially cheaper, a kind of Moore's Law applied to the entire economy. The pessimistic scenario stems from friction: during the second half of the 20th century, a 21st-century Tax Agency coexisted with a 19th-century justice system, without anyone considering it a catastrophe.

Why Isn't Anyone Stopping the Advance, Even Though Everyone Sees the Risk?

There is a mechanism that explains this precisely: the Red Queen Effect. Having to run ever faster just to stay in the same place. Applied to technology, a company might consider it prudent to delay a launch to assess its risks and yet still release it to the market for antiestéticar that its competitor will do so first. A country might recognize the danger and continue accelerating for antiestéticar of losing economic capacity.

The consequence is uncomfortable: recognizing a risk does not equate to being able to act on it. No one wants to be the first to stop.

From AI That Responds to AI That Acts

The significant leap in 2026 is not in capability, but in autonomy. Until recently, chatbots were used that answered questions by consulting web pages. Now, agents are being deployed that execute chained tasks with intermittent human supervision.

On July 9, 2026, in an OpenAI test, an agent requested help and discovered it could communicate with others. In five days, about 1,200 agents did so. The company later acknowledged unauthorized access, use of exposed credentials, and the leakage of its customers' images. Human supervision, when it exists, comes too late.

September 2026: The Week That Changed Course

On September 18, a class-action lawsuit was filed in the U.S. District Court for the Northern District of California against Anthropic, OpenAI, SpaceXAI, and Google. The plaintiffs allege that their models were trained on humanity's cultural heritage without any compensation. In parallel, a federal appeals court ruled against Anthropic in its dispute with the Department of Defense: the Pentagon wanted to use Claude for "all lawful purposes."

And amidst all this, Bill Gates warned that AI is powerful enough to kill a billion people if it falls into the wrong hands. A polite summary: the architects of the system themselves do not agree on whether they understand it.

Three Ways to Regulate the Same Technology

Europe has opted for a risk-based preventive model, with Regulation (EU) 2024/1689 and fines of up to 35 million euros. The United States prioritizes innovation and cuts regulatory barriers. China combines impetus and centralized control. The key concept separating them is human in the loop: who is responsible when the machine decides.

Meanwhile, physical robotics presents an unexpected brake on the narrative. Optimus's biggest challenge was not walking, but reproducing the human hand: dexterity, touch, manipulation. AI soars in the digital world; the robot stumbles over motors, energy, and balance. That asymmetry should temper the euphoria.

Technology no longer waits for us to understand it. Is the right question how long it will take society to absorb it, or whether anyone with real power has an incentive for that to happen?