O'Reilly Survey Reveals Wide Gap Between AI Security Threats and Organizational Preparedness

New research finds that while AI-generated attacks are accelerating, most security professionals lack specialized tools, comprehensive policies, and adequate training to defend against them

BOSTON--(BUSINESS WIRE)--O'Reilly, a knowledge company delivering the expertise that drives business outcomes, today released the findings of a new survey examining the intersection of artificial intelligence and cybersecurity. The survey, conducted in September among information security professionals, revealed a field under pressure: AI has permanently changed the threat landscape, yet many organizations remain underprepared in tools, policies, and skills.

The research captures a profession at a critical inflection point. AI-generated attacks are no longer theoretical—55% of respondents report encountering AI-generated phishing, 30% have observed AI-assisted vulnerability scanning and exploitation, and 26% have seen AI-assisted malware generation. Yet 52% of security professionals say their work rarely or never involves AI-generated attacks, suggesting that significant portions of the security workforce remain disconnected from frontline AI threats.

"The attackers aren't waiting for organizations to catch up," said Mike Loukides, vice president of content strategy at O'Reilly. "AI lowers the cost and raises the scale of attacks in ways that were hard to imagine just a few years ago. The security profession needs to move faster on tools, training, and policy to keep pace."

Key findings include:

• Concern outpaces preparedness. The top concerns among respondents were overreliance on AI tools without human verification and sensitive data leakage via AI tools, each cited by 49% of respondents. Yet fewer than half (44%) reported that their organization has a comprehensive AI security policy in place.
• Shadow AI is pervasive. 59% of respondents believe their teammates use AI outside of official policy—either occasionally (36%) or on a widespread basis (23%). Organizations with comprehensive AI policies were significantly less likely to report widespread shadow AI use (17%) compared to those with basic or no policies (35%–37%).
• Specialized tools remain underutilized. While 68% of security professionals use general-purpose AI tools like chatbots or copilots, far fewer have adopted tools purpose-built for security work: only 38% use AI for vulnerability scanning, 27% for threat detection, and 25% for incident response. Those who do use specialized tools report dramatically better outcomes—42% report improved alert triage, compared to just 18% of those using only general-purpose AI.
• Funding is improving, but unevenly. 32% of respondents say AI security is sufficiently funded at their organization, and another 28% say it is partially funded. However, organizations with comprehensive AI policies are far more likely to report adequate funding: 49% of that group say security is sufficiently funded, compared to just 16%–22% of those with weaker or no policies.
• The skills gap is real and widely felt. Only 7% of respondents said they feel no skills gap. 34% want to use AI tools more effectively, 24% want to learn how to use AI to secure systems, and 19% want to learn AI-specific attack techniques.
• Emerging threats are underestimated. Despite OWASP recognizing prompt injection as the greatest security risk for AI users in 2025, only 11% of respondents listed it as a top concern. Data poisoning was cited by just 4%, even as organizations increasingly adopt locally specialized models that raise exposure to this risk.

The survey also examined concerns specific to AI agents, whose adoption surged following the launch of OpenClaw and Gas Town in early 2026. The top concerns included excessive permissions or unclear ownership (28%), lack of monitoring or auditability (19%), and the inability to verify whether an agent stayed within its authorized role (18%).

The full survey report is available at https://www.oreilly.com/radar/the-state-of-ai-security-threats-tools-and-skills-gaps/.

About O'Reilly

O'Reilly gives teams the knowledge and judgment to determine what work matters and then go do it well. Trusted by professionals for more than 45 years, we publish the insights of the engineers, architects, specialists, and founders creating the technologies the world runs on. They explain how things actually work in the field and, more importantly, what they've learned making it happen. We deliver this expertise wherever teams need it, through the O'Reilly learning platform, Expert Intelligence for AI tools and agents, and custom programs to apply it across the business. Every format serves the same purpose: helping people and organizations build skills, make better decisions, and put knowledge into action. Learn more at oreilly.com.