Artificial intelligence systems are transforming how software vulnerabilities are discovered. Credit: Hailshadow/Getty
Recent disruptions affecting systems run by Microsoft, US air-traffic control, JPMorgan Chase and the US National Security Agency have renewed concerns about the resilience of critical digital infrastructure. Separately, researchers have used artificial intelligence to uncover security flaws in Java and PHP software, while hackers have used AI tools to target refrigerators and freezers in European supermarkets. Although the incidents are probably unrelated, together they illustrate how AI is changing cybersecurity by allowing attackers and defenders to examine vast amounts of code at unprecedented speed.
“I think vulnerability discovery without AI is basically over,” said Stephen Murdoch, a computer security researcher at University College London. Researchers who are not using the technology risk falling behind, he says.
AI can identify software weaknesses much faster than people can, potentially giving attackers time to exploit flaws before organizations patch them. Researchers are now trying to determine how thoroughly AI can assess software for security problems and whether its use will create new weaknesses.
How will AI change cyberattacks?
AI agent systems allow attackers to find vulnerabilities more quickly and systematically. In the past, hackers might have spent hours examining websites and looking for weaknesses that could provide unauthorized access to networks. AI agents can use automated tools to search millions of lines of code.
Google reports that attackers are already using agentic AI to scan software updates, identify related security issues and develop ways to exploit them.
News: Cyber attacks hit research institutions, with devastating impact
“For me, it’s a question of scale,” said Daniel Gardam, a cybersecurity researcher at the University of Surrey. AI generally finds the same types of weaknesses as human hackers, he says, but it also makes it easier for people with limited technical expertise to attack systems.
“You can launch an AI agent and come back within 20 minutes and already have dozens of results,” Gardam said.
Even when AI is involved, most security breaches still result from human error. Common examples include weak passwords and clicking links from unfamiliar sources. In European supermarkets, some targeted freezers reportedly used default passwords such as “admin” or “12345”. AI-generated messages can make phishing attempts more convincing, while publicly available information can be used to personalize fraudulent communications.
Murdoch said AI is especially effective at finding well-known categories of software flaws, although the most skilled human researchers remain better at identifying entirely new types of weaknesses. “They don’t need to sleep,” he said. “It’s no problem to work on 100 pieces in parallel.”
These capabilities helped researchers identify security flaws in Java and PHP, programming languages used to build cross-platform software and websites. In Java, for example, AI identified situations in which an attacker could corrupt a computer’s memory. Murdoch said, however, that the significance of the issue has been overstated.
Java vulnerabilities are “a real problem, but they’re not one of the top 10 vulnerabilities of the year, or perhaps this week,” he said.
Can AI help defend computer systems?
Cybersecurity researchers are using AI tools, much like hackers are, to find weaknesses in computer systems more quickly and efficiently. Murdoch said this is becoming increasingly important because vulnerability researchers can examine far more software and identify where defenses need to be patched. Researchers are also developing AI agents that can automatically detect and remediate vulnerabilities.
When security researchers find a weakness, they typically notify software developers privately. This gives engineers time to prepare a fix before the details become public.
AI is about to completely transform cybersecurity — here’s how researchers should prepare
Murdoch said the use of AI in cybersecurity has shortened the time between disclosure of a vulnerability and attempted exploitation from weeks to only days.
Rapidly issuing updates can create problems of its own. Software patches can interfere with other programs and cause outages—the very disruptions organizations are trying to prevent.
“Life has become significantly more difficult for defenders trying to solve problems,” Murdoch said.
Are the incidents connected?
Murdoch said the recent technology failures were unlikely to be linked to large-scale cyberattacks. The most notable disruption, he said, was probably caused by a technical malfunction rather than by something more malicious.
Source: www.nature.com


