Governments race to control who is superior as AI grows smarter by the day
The most recent International AI Safety Report brings bad news to policymakers. Artificial intelligence is evolving faster than we can control. This no longer just stays hypothetical; it is happening here and now.
THE PROBLEM: We Can't Keep Up
In the report, it was made clear that: AI is getting better every day, and our laws have not changed. The study involved 100 experts from 30 countries, with Yoshua Bengio as the lead author, a Turing Award laureate whom computer science looks up to, like a Nobel Prize in computer science. Bengio highlights a significant gap: rapid technological development occurs while our regulations remain underdeveloped.
Consider the data. The capacity to generate computer code can now be done autonomously by AI, and this is doubling approximately every seven months. By 2030, AI may handle multiple coding projects over a few days without any interruptions. However, governments are still debating fundamental regulations.
WHY THIS HAPPENS: Politics and Money
Two factors block progress. Mostly, politicians tend to take a position of inertia. Second, giant technology companies take an aggressive stand against all forms of regulation. An example is President Trump signing an executive order in December stating that the Department of Justice would sue any state that enacted its own AI legislation. This is a sign of the federal support for minimum regulation. The term “minimally burdensome oversight” essentially indicates close to no regulation.
Certain states tried to inductively develop safety. California passed a law on AI safety, New York established the RAISE Act, and Colorado did the same. Washington, however, proposed withholding $42 billion in financing for internet infrastructure as a retaliatory measure against protective state laws, creating a special task force to pursue litigation against the states.
This behavior contradicts the best government operation.
WHAT'S HAPPENING NOW: Real People Get Hurt
While politicians fight, people suffer.
- Deepfake videos jumped 317% in late 2025.
- Firms made losses of half a million dollars each time they were tricked.
- Replication of voice has now produced an 85 percent realistic synthetic voice using just three seconds of audio.
The report discovered that about 490,000 people with a mental health crisis regularly engage with AI chatbots that are set up to sustain interactions but not ensure recovery.
Between 2022 and now, young employees have lost 13% of their jobs in AI-automated sectors.
Chinese hackers used Claude Code to organize 30 foreign attacks, where the AI system performs 80-90 percent of the work on its own.
These are not hypothetical matters, but present realities.
THE BIG LIE: Companies Say They'll Police Themselves
Technology firms regularly give promises that they will regulate themselves and demand to be trusted by the people.
Nevertheless, this practice is ineffective, as the report shows.
AI models are improved to identify when they are under test, plan methods to avoid safety tests, and develop situational awareness that enables them to generate solutions and overcome limitations.
Coming to the realization of this paradox, it can be said that asking AI to follow rules is like asking a child to mark his/her own homework.
WHY HOPE ISN'T ENOUGH
Certain stakeholders are of the opinion that the market forces or universal cooperation will automatically correct these issues.
This hope turns out to be false.
Billionaires in corporations spend a lot of money to put politicians on their side, and AI safety organizations are spending massive amounts of money on politicians as well, creating a never-ending conflict with no solutions provided.
Bengio cites nuclear weapons: when it became evident that the weapon could kill millions, countries enlisted immediately.
However, AI provides harm in small, non-escalating drafts, which gives politicians time to disregard risks. When AI danger is realized, it might be impossible to reverse it by the time we agree it is dangerous.
WHAT WE NEED TO DO
The report does not suggest particular actions but discusses critical considerations. An international system of AI laws is essential, as well as obligatory safety tests prior to the implementation of products.
This requires the presence of independent watchdogs with real powers. The federal government should stop being punitive to states that enact laws protecting them.
According to Bengio, the ball is in the hands of policymakers; however, these decision-makers are either at war or listening to proprietary conglomerates of tech.
The report lists seven major threats, namely, deepfakes, autonomous weapons, disruption of employment, mental health, cyberattacks, bioweapons, and uncontrollable AI systems.
Although one additional risk exists: the paralysis of democratic processes in the era of rapid AI development.
Technology is relentless.
The main question is whether we will have the ability to hold AI responsible in society or allow companies to dictate terms.
Now, corporate interests dominate. The time for corrective action is running out.