Proofpoint Survey: AI Security Measures Progress, but Half of Organizations with Measures in Place Experience AI-Related Incidents

87% of organizations have moved AI assistants from pilot to production

94% of organizations report that managing multiple security tools is a challenge as AI risks spread across email, cloud, collaboration environments, and AI systems

June 11, 2026 (Tokyo) — Proofpoint Japan, a leading cybersecurity and compliance company, today announced the release of the Japanese version of the '2026 AI and Human Risk Landscape' report. This report reveals the growing gap between the rapid adoption of AI in business operations and the measures and investigation capabilities to address associated risks. This global survey of over 1,400 security professionals across 12 countries examines how rapid AI adoption is transforming collaboration within enterprises and exposing structural weaknesses in security measures and incident response.

AI is rapidly permeating organizations, moving into production across many business areas including customer support, internal messaging, email workflows, and third-party collaboration. 87% of organizations worldwide have moved AI assistants from pilot to production, and 76% are piloting or deploying autonomous agents. However, despite increased investment in AI tools and security measures, many organizations cannot confirm the effectiveness of these measures. 52% are 'not confident' that their AI security measures can detect compromised AI, and half of organizations that have already implemented measures have experienced AI-related incidents (including confirmed or suspected cases).

Furthermore, only 32.6% of organizations reported having 'sufficient investigation capabilities' for AI-related incidents spanning multiple systems and collaboration channels, revealing that many organizations are underprepared.

Survey Methodology

The '2026 AI and Human Risk Landscape' report provides a global perspective on how organizations are adopting AI and managing associated security risks. The survey examined AI adoption maturity, effectiveness of measures, incident occurrence, risks in collaboration channels, and investigation readiness as AI assistants and autonomous agents are integrated into enterprise workflows. An online survey was conducted in January 2026 with 1,453 full-time security professionals across 20 industries in 12 countries. Countries surveyed include the United States, United Kingdom, France, Germany, Italy, Spain, United Arab Emirates, Australia, Japan, Singapore, India, and Brazil.

Ryan Kalember, Chief Strategy Officer at Proofpoint, commented: 'This survey reveals a widening gap between the speed of AI adoption and security preparedness. Organizations are expanding their use of AI assistants and autonomous agents across core workflows, but many lack confidence in the effectiveness of their security measures and are unable to adequately investigate incidents that occur across collaboration channels. As AI becomes embedded in the fabric of business operations, security leaders need to rethink how to protect trusted interactions between people, data, and AI systems.'

Key Findings for Japan:

AI Adoption Outpacing Security Preparedness

AI adoption is moving into production faster than governance frameworks can mature. 84% of Japanese organizations (global average: 87%) have moved AI assistants from pilot to production, and 65% are deploying autonomous agents. Meanwhile, 57% describe their security measures as 'lagging,' 'inconsistent,' or 'reactive.' Furthermore, 47% of Japanese organizations (global average: 42%) reported experiencing 'AI-related incidents, including suspected cases,' indicating that risks in production environments are already materializing.

Collaboration Channels Become Primary Attack Surface in the AI Era

AI expands the attack surface, spreading threats at speeds far exceeding human capabilities and impacting interconnected workflows. In Japan, email remains the most common threat vector at 60% (global average: 63%), but risks are spreading across multiple channels, including third-party SaaS and cloud applications (45%), AI assistants or agents (42%), and file-sharing platforms (42%). For Japanese organizations that have experienced AI-related incidents, risks are even higher across all channels, with email at 70% (global average: 67%) and cases involving AI systems at 60% (global average: 53%).

Measures Advance, but Confidence in Effectiveness Lags

While many Japanese organizations have implemented security measures, confidence in their effectiveness is lacking. 56% of Japanese organizations (global average: 63%) report having 'AI security measures in place,' while 75% (global average: 52%) are 'not confident they can detect compromised AI.' Additionally, 61% of organizations with measures in place still reported experiencing 'AI-related incidents.' In Japan, challenges include governance alignment across teams (58%), training (44%), and visibility into AI or agent activity (43%), indicating gaps in response capabilities.

Incident Reality Outpaces Investigation Preparedness

When AI-related incidents occur, many Japanese organizations struggle to investigate effectively. Only 16% of organizations (global average: 33%) reported being 'well-prepared to investigate' AI or agent-related incidents, and 45% (global average: 41%) said they 'find it difficult to correlate threats across channels.' As AI-related activities span multiple environments like email, collaboration platforms, and cloud systems, visibility across the interconnected environment is essential for event reconstruction, but many organizations still lack sufficient visibility.

Tool Proliferation Creates Structural Barriers

Fragmentation of the security stack exacerbates challenges, impairing visibility and delaying response as incidents move between systems at speeds far exceeding human capabilities. 92% of Japanese organizations (global average: 94%) find managing multiple security tools at least 'somewhat difficult,' with 41% (global average: 52%) finding it 'very or extremely difficult.' Key drivers include operational cost pressure (53%), difficulty in threat correlation (45%), and tool overlap or redundancy.