Fast AI, Slow Safety

Something unusual has happened in the world of artificial intelligence this month. The very architects racing to build faster, more powerful systems are stepping on the brakes and issuing a collective warning: it might be time to pause.

Anthropic CEO Dario Amodei recently sounded the alarm on the breakneck pace of AI development. His core message is simple: as technology races ahead, are our safety preparations keeping up?

He isnot alone. OpenAI’s Sam Altman, Google DeepMind’s DemisHassabis, and Microsoft’s Satya Nadellahave voiced identical concerns.

Thisdefinitelymatters because it isnot outside critics. They are the ones building this technology and shaping its future.

When people competing for the same customers, investment, and talent start saying the pace itself needs attention, that’s hard to ignore.

Where the Game Is Changing
Why the sudden anxiety? The game is shifting rapidly.

Amodei warns that within the next six to twelve months, AI capabilities could advance to the point where hundreds or thousands of AI agents working together could orchestrate large-scale cyberattacks.

This sounds like science fiction, but AI has evolved past simply answering prompts.

Modern systems can plan, execute software tasks, and gather information sequentially-drastically reducing the need for human hand-holding.

That is exactly where the opportunity and the risk both live.

Imagine an AI given access to a company’s software, financial data, or supply chain.

It can work fast-that’s the upside. But if it makes a wrong decision, that mistake spreads just as fast.

A chatbot giving a wrong answer is one kind of problem; an autonomous AI making a wrong call that damages a company’s money, data, or operations is something else entirely.

Worse yet is the concept of “recursive self-improvement”-a scenario where AI begins playing a major role in upgrading its own code.

If realized, the pace of progress could hyper-accelerate, leaving humans struggling to understand how decisions are made or where to draw the line.

The Myth of the Smart Model
One common misconception about AI is that a more powerful model means a more accurate one.

That’s not true. An AI’s output depends on the data it’s given, how the model is designed, the instructions or prompts it receives, and the context available to it. Human judgment still matters enormously.

If the data is wrong, incomplete, outdated, or biased, the AI’s decisions can be weak as a result.

Generative AI can also produce information that sounds entirely convincing but is factually wrong-we’ve already seen examples of fabricated statistics, invented citations, and information that simply doesn’t exist.

A lack of context can cause problems too.

Say an AI analyzes several years of sales data and recommends increasing production.

But it doesn’t know that the company’s biggest client just cancelled their order. The math may be correct, but the decision is wrong.

In other words, no matter how intelligent an AI is, the quality of its decisions depends on what it has been taught and how carefully humans verify its output.

A Word of Caution for Bangladesh
If we want to adopt advanced AI, buying good software isn’t enough.

We need reliable data, strong cybersecurity, a skilled workforce, and governance structures where humans remain accountable for major decisions.

In many of our institutions, data is still scattered and disorganized. One department uses one format, another uses a different one.

Some records are in Bangla, others in English. Much of the data is outdated or incomplete.

So, our challenge isn’t just a shortage of data-it’s a shortage of high-quality, reliable, well-governed data.

Making large investments in AI without first fixing this foundation could backfire. Bad data would simply get analyzed faster, and wrong decisions would be made with even greater confidence.

So, the real question facing Bangladesh isn’t whether to adopt AI. It’s how to adopt it in a way that captures the benefits while keeping the risks under control.

Neither Rushing in Nor Falling Behind
A sensible path for Bangladesh would be to move quickly wherever AI can boost productivity with limited risk-but to proceed carefully in sensitive areas like people’s livelihoods, financial security, personal data, and national security.

The government should establish a clear policy framework for AI use.

This could include forming a national AI safety and governance council, bringing together experts in technology, industry, cybersecurity, law, economics, and education.

This council’s job would be to determine where risks are highest, where direct human oversight is needed, how data should be protected, and who is accountable when something goes seriously wrong.

Businesses should ask themselves similar questions before adopting AI: Is our data reliable? Are our employees ready to use this technology? Is customer data protected? Will humans remain in control of important decisions?

Before asking “Where can we use AI?”, organizations should first ask: “What problem do we actually need to solve, and is AI the safest and most effective way to solve it?”

The bottom Line
The warnings from AI leaders are not a rejection of AI; they are a reminder that building technological capability and using it responsibly are two different things.

For Bangladesh, the goal should not be to blindly chase the global AI race, but to strengthen our data, infrastructure, skilled workforce, and governance first. The better prepared we are, the more we can benefit from AI.

The real question is not how fast AI advances, but how prepared we are to move alongside it.

There is no need to fear AI-or to trust it blindly. How, where, and how much we use it is ultimately our choice.

(The writer is distinguished Professor, Eastern University, andFormer Vice Chancellor, East West University, Bangladesh)