A new Linux Foundation report finds that 58% of software development teams are releasing code faster than their security staff can review it, raising major concerns about vulnerabilities introduced by AI coding tools.
The report, State of AI Impact on Software Security and Cybersecurity 2026, surveyed 308 cybersecurity practitioners working in businesses and open source projects. The research reveals troubling gaps between the rapid adoption of AI development tools and the ability of security teams to manage the resulting risks.
The problem isn’t AI itself. The issue is that many companies lack the staffing and security processes needed to deploy AI safely.
A particularly troubling finding involves the software supply chain. Some 60% of respondents reported that AI coding tools add external software dependencies to their applications, compared with just 16% who reported fewer dependencies. These third-party software components can contain vulnerabilities that expose applications to cyberattacks. Yet development teams review only 53% of indirect dependencies, leaving much of their software infrastructure without adequate scrutiny.
Among the survey’s more experienced AI cybersecurity pros, 93% expressed concern about software supply chain attacks, followed by prompt injection attacks at 90% and the maintenance demands created by frequent software updates at 87%.
The research also found that 76% of experienced cybersecurity practitioners believe AI makes it easier for attackers to combine multiple vulnerabilities into more sophisticated exploits.
The Value of Oversight
Despite these threats, relatively few organizations have established dedicated oversight for AI security. A mere 30% of surveyed teams have assigned specialized personnel to handle AI-related cybersecurity threats. The figure rises to 35% among businesses but falls to 16% for open source projects.
Formal governance is limited. About one-third of organizations have adopted official security policies addressing AI, while 35% rely on informal guidelines.
Even where policies exist, security requirements vary considerably. While 76% identify approved AI tools, only 57% require human review of AI-generated code. Just 36% include provisions for automated security scanning.
The lack of oversight, of course, has major consequences. Among teams with dedicated AI security personnel, 43% investigate and resolve most new security alerts immediately. That figure drops to 23% among teams without dedicated AI security resources.
On a positive note, AI is proving valuable as a defensive technology. Practitioners credited AI-assisted security tools with identifying approximately 30% of the vulnerabilities they discovered during the previous year. Some 69% of practitioners using these tools reported meaningful improvements in cybersecurity, while 72% viewed AI as a supplement to existing security practices.
AI security tools helped practitioners complete security-related development tasks 17% faster, on average. But inaccurate vulnerability alerts generated by the tools added an estimated 30% to workloads, forcing security teams to devote additional resources to evaluating potential threats.
Another concern involves developer expertise. Some 82% of respondents agreed that excessive dependence on AI tools weakens developers’ fundamental security skills.
Bottom line: AI can speed up development, but greater oversight is definitely needed. The Linux Foundation recommends stronger controls over third-party software, formal AI security policies, continued developer training and greater use of automated security testing.

