Spirantix.aiEducational Non-Profit

Lesson 6 · 11 minute read

Is AI actually dangerous?

What the news is really reporting about AI risk, data centers, cyberattacks, and calls for regulation.

Three questions come up often when AI is in the news. Can the companies building these systems actually control them? Is AI about to get out of hand? And can elected leaders do anything about it?

No one, not the AI companies, not the researchers who study them, not the lawmakers who regulate them, has a settled answer. That is the honest starting point. This lesson looks at what has actually happened in the news, separates it from what might happen, and gives you the sources so you can check for yourself.

Why this has been in the news

On September 8, 2026, a researcher named Jacob Coxon posted that he had resigned from Anthropic, one of the companies that builds AI assistants. He said the leading AI companies were racing ahead faster than they could keep the technology safe.

What made this a bigger story is who agreed with him publicly, under their own names. A researcher who monitors AI systems at OpenAI put the odds of a very bad outcome, without stronger safety rules, at 70 percent. A former senior safety researcher who worked at OpenAI, DeepMind, and the UK government's AI safety office estimated a 50 percent chance of serious harm within the next decade. Anthropic's own alignment lead said he believes the risk is real.

These are not outside critics. They are, or were, people working inside the companies building the technology, which is why the story traveled the way it did. It does not prove any one number is correct. Estimates like these cannot be tested the way a weather forecast can. What is verifiable is that senior people inside the leading AI companies are now saying this in public.

Worth knowing: Concern about advanced AI is not new. A nonprofit called the Machine Intelligence Research Institute has argued for over a decade that AI poses a serious risk, well before this news cycle.

The environmental question

You may have heard AI data centers compared to a desert city watering its lawns: using water a region cannot spare. The real picture is more mixed than that.

Why a data center needs water at all

Computer equipment generates heat, and heat has to go somewhere. Many data centers cool themselves the way your body cools itself when you sweat: water absorbs heat and evaporates, carrying the heat away. It works well, but it uses a lot of water. The International Energy Agency estimates that a large data center can use roughly half a million gallons of water a day, about as much as 6,500 homes. Not every data center uses this much. Newer designs use closed-loop or air cooling that recycles the same water instead of evaporating it, which some companies are now building specifically to avoid this problem.

Data centers are a small slice of Arizona's water use today. That slice is growing fast and is not tracked well, which is the honest complication.

Is this only an Arizona problem?

No. It shows up wherever a data center is built in a place that is already short on water. In Georgia, a drought and a fast-growing cluster of data centers around Atlanta led planners to ask developers to cut back, with water requests for individual projects ranging from about 5,000 gallons a day for efficient closed-loop systems up to 9 million gallons a day for older designs. In central Iowa, a water use ban in 2025 led some people to blame Microsoft's local data centers, but the utility that manages the water supply said the real cause was contamination in two rivers that limited how much could be treated, and that lawn watering used far more water than the data centers did once the ban took effect.

Outside the US, it has become a bigger flashpoint. A Chilean court ordered Google to redo its environmental review for a data center project in Santiago after residents objected to a plan that would have used about 2 million gallons of drinking water a day during a drought; Google has since offered to switch to air cooling instead. Data center water use has also drawn public protests in Uruguay and the Netherlands, and both China and India are seeing a larger share of new data centers built in drier regions.

Is anyone making the case this is overblown?

Yes, and it is worth reading. A July 2026 report from the Information Technology and Innovation Foundation argued that data centers use less than 1 percent of total water consumption in the United States, that there is no nationwide shortage, and that the real problems are local, in already-dry places like Arizona, rather than a general crisis. It also pointed to newer "zero water" cooling designs that some companies are already building. The Iowa case above is a real example of a local water problem that data centers were blamed for and, on closer look, mostly were not responsible for.

Put together, the honest summary is: this is a real and growing issue in specific dry places, not evidence that data centers are draining water everywhere, and the industry has a clear technical path to using much less of it.

On energy, the trend is genuinely improving. New chip designs and cooling methods published in 2026 could cut AI energy use and improve efficiency significantly, though most of this is still moving from the lab toward real-world use. Total AI energy use is still rising as more data centers get built, but each individual task is getting less wasteful, not more.

What has actually caused damage

Set aside hypothetical future harms and look at what has actually hurt people and businesses. The honest answer: ordinary cybercrime, most of it not using AI at all.

Clorox, the consumer products company, lost an estimated $356 million after attackers tricked its IT help desk into resetting an employee's password in 2023. A UK transport company founded in 1865, employing 700 people, was forced to close in 2025 after a ransomware attack that started with one employee's weak, easily-guessed password.

Neither of those required a misbehaving AI system. Both required one weak point in ordinary account security.

Important: The single habit that would have stopped both of those attacks is one you already use in earlier lessons: strong, unique passwords and multi-factor authentication wherever it is offered.

What AI actually is

It helps to be precise about what a system like ChatGPT, Claude, or the Spirantix Concierge actually is. It is a predictive system, guided at every step by the instructions a person built into it. It does not want anything on its own, and it is not self-aware the way a person is.

That does not mean it cannot be misused. It can, the same way any powerful tool can. But the tool is not what is responsible for the misuse, any more than a car is responsible when someone drives it recklessly. The line between a tool and the person using it is worth holding onto.

The regulation question, and who benefits

The leading AI labs have started asking government to set safety rules: independent oversight, common safety standards, and coordination between companies. Several well-known industry leaders have backed this idea.

Here is the part worth sitting with. A small number of companies, already far ahead of everyone else, are asking government to set rules that mainly companies at their scale can easily meet. Critics have pointed out this can work like a competitive shield: it can be sincerely meant as a safety measure, and it can also make it harder for a new competitor to catch up. Both things can be true about the same proposal.

Quick check: does a real safety worry rule out a business motive?

No. A company can hold a sincere safety concern and a competitive interest in the same proposal at the same time. Reading one does not require dismissing the other. The useful habit is asking who benefits from a proposed rule, not just what the rule claims to do.

What this means for you

None of this is a reason to panic, and none of it is a reason to stop paying attention either. A few habits hold up regardless of how the bigger questions get resolved:

  • Treat AI tools the way you would treat any powerful tool operated by a person: useful, worth understanding, and worth double-checking on anything that matters.
  • Keep a human decision in the loop for anything involving money, health, or a legal matter.
  • Practice the same account security habits covered in earlier lessons. That single habit stops nearly all the real-world damage described above.
  • Stay skeptical of both extremes: the version that says AI will end the world tomorrow, and the version that says none of this is worth a second thought.

Try this today

Ask ChatGPT or Claude:

"Explain, in plain language, the difference between a real AI safety risk and a company's business interest in regulation. Give me one question I should ask when I read a news story about AI danger."

Read the original reporting

Every claim above was checked against news reporting from September 2026 at the time this lesson was written. These are some of the original stories, so you can read further and judge for yourself:

A full numbered reference list with every source used for this lesson is available on request.

Last reviewed: September 17, 2026

This lesson summarizes reporting from multiple news outlets, listed above, current as of September 17, 2026, including sources on both sides of the data center water debate. Two examples originally considered could not be verified and were replaced with confirmed cases; see the full reference list for details.

AI products change frequently. If a button or menu looks different, use the product’s Help area or official documentation for the newest instructions.

Back to the Learning hubSee what is coming next