SandboxAQ Report Highlights AI Security Gaps Amid Rapid Adoption

SandboxAQ has released its first AI Security Benchmark Report, revealing a stark contrast between the widespread adoption of AI and the implementation of security measures. According to the report, while 79% of organizations are using AI in production, only 6% have adopted a comprehensive AI-native security strategy, leaving many enterprises vulnerable to AI-related threats announced in a press release.

The report, based on a survey of over 100 senior security leaders in the US and EU, highlights concerns about risks such as model manipulation, data leakage, and adversarial attacks. Despite these concerns, only 28% of organizations have conducted a full AI-specific security assessment, and most rely on traditional security tools not designed for AI systems.

The rise of non-human identities (NHIs), including autonomous AI agents and machine accounts, further complicates the security landscape. These entities often operate independently, holding cryptographic credentials and accessing sensitive resources without human oversight. The report notes that most security teams lack visibility and control over these NHIs, undermining core security principles.

Despite these challenges, investment in AI security is on the rise, with 85% of organizations planning to increase spending in the next 12 to 24 months. Key focus areas include protecting training data, securing NHIs, and deploying automated incident response capabilities tailored to AI infrastructure.