They are often trained on public domain code, which can be secure or insecure. The AI coding assistant is not able to identify which is which. It also rewrites code from these sources without noticing any logical issues that might exist. AI is rewarded based on whether it completes a task, not if it is done well, so it might create code that is not secure or without necessary security controls.
That's according to an analysis by Bugcrowd, which found the last year has seen a massive 88% increase in hardware vulnerabilities and a doubling in network flaws. In a new report, Inside the Mind of a CISO 2025: Resilience in an AI-Accelerated World, the firm said that 81% of security researchers had encountered new hardware vulnerabilities in the past 12 months.