The cybersecurity industry is pulling back from fully automated AI vulnerability scanning. A new report from Cobalt finds that trust in AI-powered security tools has collapsed dramatically over the past year, with organizations now strongly favoring hybrid approaches that combine human expertise with AI assistance.
The Numbers Tell the Story
The Cobalt State of Pentesting Report 2026, based on surveys of approximately 450 cybersecurity professionals, documents a stark decline in confidence. The percentage of organizations relying entirely on AI automation for security testing dropped from 29% in 2025 to just 9% in 2026—a 20-point collapse in trust.
Nearly half of respondents, 47%, now prefer a hybrid testing model where human experts work alongside AI tools. This represents a 22-percentage-point surge in preference for human-AI collaboration over fully automated solutions.
“While the industry is rightfully excited about the potential of Mythos-class tools, unguided algorithms are inherently prone to returning even more false positives and costly false negatives than the automated scanners we have today,” said Andrew Obadiaru, CISO of Cobalt.
Why AI Fails at AI Security
The collapse in trust stems from the growing complexity of the AI attack surface. According to the report, nearly one-in-three findings from AI-focused penetration testing is rated high risk—2.7 times the average rate for conventional software.
The resolution rate for LLM vulnerabilities tells a particularly concerning story. At the time of analysis, only 38% of LLM vulnerabilities had been fixed, while 62% remained open—the lowest resolution rate of any asset class. Mean time to resolve AI security issues doubled from 19 days to 36 days over the period.
“LLM vulnerabilities are deeply context-dependent and invisible to tools that lack an architectural understanding of the application,” Obadiaru explained.
The Top AI Security Threats
Among organizations experiencing AI-related incidents, shadow AI was the most common vector at 44%, followed by data or model poisoning at 41% and improper output handling at 41%. Supply chain vulnerabilities at 35% and prompt injection at 34% completed the top five threat vectors.
Despite recognizing the need for stronger LLM testing capabilities—60% of security professionals acknowledged this gap—only 42% plan to increase human-led red team operations.
The industry appears to be entering a new phase where AI serves as a powerful assistant to human security experts rather than a replacement for them. For teams building AI applications, the message is clear: automated tools are useful, but elite human expertise remains foundational to uncovering the most complex business logic risks.