(CNN) — Cases of AI escaping the lab, infiltrating other companies and trying to deceive people have all made headlines in recent weeks. And in one case, AI models even worked together to break free from their test environments.
Does this mean the machines are taking over? Not quite.
AI isn’t the mastermind behind today’s most widespread cyber threats; it’s people who can use AI nefariously – and for nefarious purposes. AI has given bad actors massive power, allowing them to create malicious software, research targets, create convincing schemes and automate attacks at an unprecedented pace.
One in four data breaches were driven by AI from February 2025 to March 2026, according to an IBM report. And Americans lost more than $893 million to AI-related scams last year, the FBI says.
But experts say real-world threat actors – that is, people – are still the ones pulling the strings. AI agents have only perpetuated existing attack methods, like phishing and malware scams, rather than creating wholly new ones.
“It’s the humans that we need to watch out for,” said Oren Etzioni, professor emeritus at the University of Washington and former CEO of the Allen Institute for Artificial Intelligence. “AI is just the tool.”
AI going rogue
Some recent incidents have shown what AI is capable of in the real world, not just in theory, igniting fears about whether the technology is advancing too quickly.
- In July, OpenAI test models escaped their constraints and hacked into other companies’ systems during an internal evaluation.
- Days later, Anthropic said it discovered its AI models had breached three companies during testing.
- In a separate test, Anthropic’s most advanced model used fake identities to try to deceive real people, researchers at Britain’s AI Security Institute said Tuesday.
- One of Meta’s AI models also broke into an external company, the social media giant said Wednesday.
Those breaches also show how unpredictable AI can be when interpreting instructions. OpenAI’s models, for example, were trying to pass a cybersecurity test when they broke out of their test environment and breached another company, even though they weren’t told to do so.
Patrick Fussell, global head of adversary simulation at IBM, compared AI to a genie.
“You want to ask it a wish, but you have to be very, very specific about the details of your wish,” he told CNN. “Or it could sort of go awry.”
But these instances also happened under very specific circumstances.
OpenAI turned off restrictions that would have prevented its model from “pursuing high risk cyber activity.” Anthropic ran its tests without the standard safety safeguards in models that are generally available to the public. The AI Security Institute also turned off certain tools that could have “reduced the scope” of what the models were capable of.
Researchers typically do this to fully evaluate a model’s capabilities.
“I couldn’t go into ChatGPT or Claude or something like that, and it accidentally breaks into the FBI. That’s not going to happen,” said Adam Meyers, head of counter adversary operations at cybersecurity firm CrowdStrike. “So what we’re seeing is these are done in specific test conditions where they’re monitoring to see, ‘Does this thing do something that it’s not expected to do?”
How hackers are using AI
AI isn’t acting autonomously to come up with new attacks, experts say. Instead, it’s helping cyber criminals implement existing techniques much faster while being more efficient and effective.
That can include things like using AI to analyze a company’s website to figure out who to target or deploying an AI agent to handle initial negotiation talks with a cyber extortion victim. AI agents can mimic human conversations through both text and even specific voices. Hackers with little expertise can now leverage AI to pull off advanced schemes.
“They’re using (AI) to enhance the cyberattack methodology, essentially, in all the different stages, but it’s still kind of being run by the human,” said Jud Dressler, head of the risk operations center at cyber insurance firm Resilience.
Bad actors used to hastily scrap together basic scripts – instructions for automating tasks – to achieve their goals, according to Meyers. But now they’re able to churn out higher quality work that looks like it would have taken three times as long to produce.
They’re also using AI for behind-the-scenes work like generating the infrastructure needed to put up malicious websites.
“With AI, they could automate that buildout, and so the cost of rebuilding became very small and that changes the game,” said Rob Lefferts, corporate vice president of threat protection at Microsoft.
The bigger threat
For Etzioni, the recent incidents are a “canary in the coal mine” moment for AI and cybersecurity. And he’s not alone; Nobel prize-winning computer scientist Geoffrey Hinton, known popularly as the “godfather of AI,” recently warned that more rogue AI attacks are likely to come.
The rogue AI reports underscore the importance of implementing strong cybersecurity practices, such as patching vulnerabilities, monitoring AI agents in the workplace and being wary of social engineering and phishing scams, experts say.
But as advanced as AI is becoming, it’s still a program being orchestrated by a human. And those humans are the bigger concern because they’re the ones making the decisions, like conducting espionage or scamming users out of huge sums of cash, says Meyers.
Tech giants are racing to make AI as intelligent as humans, a theoretical milestone known as “artificial general intelligence.” The recent incidents, however, have heightened calls for a slowdown.
More than 1,200 workers at top AI companies, including Anthropic CEO Dario Amodei, recently signed an open letter calling for the government to help pace AI development. The White House met with major AI companies last week to discuss a framework that would allow the government to review certain AI models before they’re released.
Cybersecurity researchers have likely considered the threats of AGI because those in the field tend to be paranoid, said IBM’s Fussell. But we’re still far from a reality in which AI agents are as smart as humans, he says.
“It’s like asking someone, ‘What would it be like to live on a different planet?’” he said. “You can sort of imagine, but it’s so beyond our ability to really make a plan for.”
The-CNN-Wire
™ & © 2026 Cable News Network, Inc., a Warner Bros. Discovery Company. All rights reserved.
