AI Gets More Powerful and the Argument About Who Should Have It Gets Louder

Key points

• Concerns about the misuse of increasingly powerful AI models are growing among researchers and policymakers.
• The worry centres on criminal use: bad actors exploiting AI to automate or scale harmful activity.
• No single incident or named group triggered the latest warnings; the concern is structural, not reactive.
• The debate has cycled before, but each new generation of models restarts it with higher stakes.

Artificial intelligence, meaning software that can write and reason on its own, is getting measurably more capable. That has revived an old argument about who gets to use it.

SecurityWeek flagged the latest round of concern this week. The specifics in the public record are limited. What security researchers and policy voices are pointing to is a broad pattern: as AI models grow more powerful, the gap between what a skilled professional can do with them and what a criminal can do narrows.

What is the actual worry here?

The concern is that AI tools lower the bar for criminal activity. A person who wants to run a fraud scheme, write convincing fake messages to trick victims, or probe computer systems for weaknesses no longer needs years of specialist knowledge. A capable AI model can help fill that gap.

This is not a hypothetical. Ransomware gangs, criminal organisations that use malicious software to lock a company's files and demand payment to restore them, already use AI-generated text to make their phishing emails, fake messages designed to trick staff into handing over passwords, look more convincing. The tools available to those groups improve every time a new model is released. On 2 September we reported that OpenAI's Astra became the first AI model to hit a landmark hacking benchmark, meaning it can hunt down and exploit unknown software flaws on its own. That capability, in criminal hands, is exactly what these warnings are about.

What is new in the current wave of concern is scale. The models being discussed now are general-purpose systems, capable of helping across many kinds of tasks, which means the range of potential misuse is wider too.

Should ordinary people care about this?

Yes, in a practical rather than abstract sense. The downstream effect of AI-assisted crime lands on real people: patients whose hospital records get locked by ransomware, customers whose payment details get harvested through a smarter phishing campaign, small businesses that lose months of data.

If you receive an unexpected email asking you to click a link or confirm account details, be sceptical regardless of how polished it looks. AI makes fake messages harder to spot.

The blunt truth is that this debate has never really been resolved, because the technology keeps moving faster than the policy designed to contain it. Watching which governments introduce binding rules, and which AI developers volunteer meaningful safety commitments, is what matters most right now.

Common questions

Is AI being used in cyberattacks right now?

Yes. Criminal groups already use AI tools to write more convincing phishing emails and automate parts of their operations. The concern raised in the latest warnings is that this will become easier and more widespread as models improve.

What can organisations do about AI-assisted threats?

Training staff to recognise suspicious messages is still one of the most effective defences, because many AI-assisted attacks still rely on tricking a person into taking an action, like clicking a link or entering a password.