
The disruption of multiple computer systems run by Microsoft, US air traffic control, the bank JPMorgan Chase and even the US National Security Agency has renewed concerns about the resilience of critical digital infrastructure.
At the same time, researchers have reported artificial intelligence -assisted discovery of security flaws in Java and PHP software, while hackers have used AI tools to target supermarket fridges and freezers in Europe.
Although probably unrelated, these incidents highlight a broader transformation in cybersecurity, say researchers. Artificial intelligence is changing how software vulnerabilities are discovered, allowing both attackers and security specialists to examine vast quantities of computer code at unprecedented speed.
“I think vulnerability discovery without AI is basically dead,” says Steven Murdoch, a computer security researcher at University College London. Researchers in the field who don’t use the technology now risk falling behind, he says.
AI can find weaknesses in software much faster than humans can, giving attackers time to exploit vulnerabilities before organizations can fix them. Researchers now want to understand how thoroughly AI can check software for security flaws, and whether it could introduce new weaknesses.
How is AI changing cyberattacks?
Agentic systems help attackers identify security vulnerabilities more rapidly and systematically. While hackers previously had to spend hours inspecting websites and probing software for weaknesses that offered unauthorized access to networks, AI agents use automated tools that can search millions of lines of computer code.
Google has reported that attackers have already used agentic AI to examine software updates, identify the security problems they address, and then find ways to exploit them.

NEWS: Cyberattacks are hitting research institutions — with devastating effects
“For me, it's about the scale,” says Daniel Gardham, a cybersecurity researcher at the University of Surrey. AI tends to find the same kinds of weaknesses as human hackers, but it makes it easier for people with limited technical expertise to break into systems.
“They can set an AI agent off and come back in 20 minutes and have dozens of results already,” he says.
Even with AI, most security breaches still involve human failure, he says, such as weak passwords (a common problem with Europe’s supermarket freezers was that passwords were defaults such as ‘admin’ or ‘12345’) or clicking on links from unfamiliar sources. AI-crafted messages can make such phishing messages more convincing and so more likely to succeed. They can draw on publicly available information to personalize fraudulent communications and make them appear authentic
Murdoch says AI is particularly good at discovering familiar kinds of software flaws, even if the most accomplished human researchers remain better at finding entirely new types of weakness. “They don't need to sleep,” he says. “You can happily work a hundred of them in parallel.”
These capabilities of AI helped discover security flaws in Java and PHP software, two programming languages that build cross-platform software and websites. In the Java case, for example, AI identified circumstances that could allow an attacker to corrupt computer memory. But Murdoch says the significance of the issue has been exaggerated.
The Java vulnerability, he says: “Is a real problem, but it’s not one of the top ten vulnerabilities this year or probably even this week.”
Can AI help defend computer systems?
Just like hackers, cybersecurity researchers are using AI tools to quickly and efficiently find weaknesses in computer systems. Murdoch says this has become increasingly important to vulnerability researchers, who can now examine much larger quantities of software and work out how to patch up the defences where necessary. Researchers are also developing AI agents that can detect and repair vulnerabilities automatically.
When security researchers uncover a weakness, they typically notify the software developer privately, allowing engineers to prepare a remedy before publishing details of the problem.

AI is set to completely transform cybersecurity — here’s how researchers must prepare
Murdoch says the use of AI in cyber security has shortened the time window between public disclosure and attempted exploitation – from several weeks, to just a few days.
Rushing an update can bring its own problems. A software patch can interfere with other programs and cause the sort of outage that organisations are trying to avoid.
“The life for the defenders who are trying to fix things has become significantly harder,” Murdoch says.
Are the incidents connected?
It’s unlikely that the recent tech outages were connected as large-scale cyber-attack, says Murdoch. The most prominent disruptions, he says, are likely technical malfunctions rather than anything more sinister.
Enjoying our latest content?
Log in or create an account to continue
- Access the most recent journalism from Nature's award-winning team
- Explore the latest features & opinion covering groundbreaking research