Hackers are increasingly using artificial intelligence to detect software vulnerabilities, reducing the time organisations have to respond to cyber threats, Verizon said in its annual data breach report.

The company said exploitation of software flaws has overtaken stolen credentials as the most common starting point for breaches for the first time. In a review of more than 31,000 incidents, Verizon found that 31 per cent of breaches began with vulnerability exploitation. The report said “AI is fundamentally reshaping the cybersecurity industry.”

According to Verizon, intruders are using generative AI across several stages of attacks, including targeting, initial access, and the development of malware and other tools. The report said AI is being used by threat actors “to accelerate the time to exploit known vulnerabilities, shrinking the window for defence from months to mere hours.”

The findings add to a series of recent warnings about the role of AI in cyber incidents. CrowdStrike said in its annual global threat report earlier this year that in 2025, “AI-enabled adversaries increased attacks by 89 per cent year-over-year.” It said the technology had strengthened less sophisticated threat actors while amplifying the capabilities of more advanced groups.

Shadow AI raises insider data risks

Verizon also said the use of shadow AI, or unauthorised AI tools, has become the third most common non-malicious insider action linked to data loss incidents. The report said employees are submitting source code, images and other structured data into such systems, creating new risks for organisations already facing faster external attacks.

Verizon said AI’s main impact is currently operational, automating and scaling techniques that defenders already know how to detect, rather than creating entirely new attack methods. However, the company warned that this assessment could quickly become outdated as AI advances.

The report does not include data from Mythos, a new AI model that has raised cybersecurity concerns. Mythos was announced on 7 April and is being deployed through Anthropic’s “Project Glasswing”, a controlled initiative allowing selected organisations, including Verizon, to use the unreleased Claude Mythos Preview model for defensive cybersecurity purposes.

Experts say Mythos’ advanced coding capabilities could give it an unprecedented ability to identify cybersecurity vulnerabilities and develop ways to exploit them.

Read more:

• Humanoid says AI brain can train robots in days as humanoid race accelerates
• Apple and Meta oppose Canadian bill over encryption and privacy concerns

Verizon chief information security officer Nasrin Rezai told Reuters that addressing the growing threat is critical. “We need to fight AI with AI. We need to incorporate them into our practices,” Rezai said. “We need to bring them into our software development life cycle, in our testing processes, in our cyber defence processes at a scale that we have never done before.”

Tags

News Artificial Intelligence cybersecurity Digital Security Machine Learning