Cybersecurity professionals lose faith in fully automated AI testing

Per Tech Radar, confidence in fully automated AI testing for cybersecurity vulnerabilities has significantly declined, with a sharp drop from 29% in 2025 to just 9% this year, according to Cobalt's 2026 State of Pentesting Report.

The report, based on surveys of approximately 450 cybersecurity professionals, indicates that 78% of respondents observed automated tools failing to detect critical vulnerabilities. The complexity of AI attack surfaces and context-dependent flaws, particularly with Large Language Models (LLMs), contributed to this decline. The mean time to resolve (MTTR) for AI/LLM security issues has risen from 19 to 36 days, with a substantial 62% of LLM vulnerabilities remaining unresolved at the time of analysis.

Consequently, hybrid testing models, which combine automation with human expertise, have seen a surge in adoption, now favored by 47% of professionals. This shift highlights the industry's recognition that while automation excels in certain areas, human expertise remains crucial for identifying and remediating complex business logic risks.

Source: Tech Radar