Executives see GenAI as a practical way to boost productivity, improve processes, and open new revenue streams, so they are pushing hard for adoption across the business.

Recent research from Microsoft Security and MDC Research Group, based on 400+ enterprise IT and security decision makers, shows how widespread this has become:

• 95% of respondents are already using or developing GenAI, or plan to.
• Only 5% say they are not using or developing GenAI and have no plans to do so.
• 66% of organizations are not just using GenAI apps, but are also developing or planning to develop their own GenAI applications.
• Among organizations building GenAI, the average number of apps in development is 13.9 – so most are working on multiple use cases, not just one pilot.

Why build or customize their own GenAI apps instead of relying only on off‑the‑shelf tools?

• 58% want to drive business innovation with customer-facing GenAI experiences.
• 55% want to maintain control of their data.
• 52% need to integrate GenAI with existing systems.
• 49% are focused on cost and scalability versus pure SaaS.
• 44% are driven by compliance and regulatory requirements.

In short, organizations are reimagining how they work and serve customers with GenAI, and they are doing it at scale. But that same speed and scale is creating a new set of security expectations for security and risk leaders.

Security and risk leaders want to say “yes” to AI, but only if they can do it safely. The research highlights a clear set of concerns that show up both when organizations use GenAI and when they develop GenAI applications.

Top concerns when using GenAI

For organizations already using GenAI, the leading worries are about data and output quality:

• Leakage of sensitive data – 63%
• Oversharing of sensitive data (users seeing data they shouldn’t) – 60%
• Inappropriate use or exposure of personal data – 55%
• Inaccurate insights generated by GenAI – 43%
• Harmful or biased outputs – 41%

These concerns are amplified by user behavior. For example, 78% of AI users bring their own AI tools to work (BYOAI), often pasting in sensitive content like source code, meeting notes, or spreadsheets. That creates a real risk of data leaving the organization without security teams knowing.

Top concerns when developing or customizing GenAI apps

When organizations build or customize GenAI applications, the risk picture expands:

• Data leakage and exfiltration – 60%
• Inappropriate use of personal data – 50%
• Violations of regulations – 42%
• Lack of visibility into AI components and vulnerabilities – 42%
• Over-permissioned access granted to AI apps – 36%

On top of these, leaders are increasingly focused on emerging AI-specific risks such as:

• Hallucinations – incorrect or misleading responses that can affect decisions in areas like healthcare, finance, or legal.
• Prompt injection attacks – malicious prompts that cause the model to ignore safeguards or reveal confidential information.
• Training data poisoning – tampering with training data to introduce backdoors, bias, or vulnerabilities.
• Model theft – stealing proprietary models, which can erode competitive advantage and create new abuse channels.
• Excessive agency – AI systems with too much autonomy performing unintended or harmful actions.

Regulation is another pressure point. Under the EU AI Act, violations can cost up to €35 million or 7% of annual turnover, yet 62% of business leaders say they do not fully understand which AI regulations apply to their sector.

All of this explains why security leaders are pushing for a more structured, security-first approach to GenAI adoption.

The research points to a clear theme: organizations want a “secure yes” to AI. They don’t want to slow innovation, but they do want a more deliberate security foundation.

In fact, 95% of security and risk leaders agree their company needs security measures in place for AI apps—including third‑party SaaS, enterprise-ready, and custom-built GenAI—within the next 12–24 months.

A practical path forward includes four key moves:

1. Form a dedicated security team for AI

Create a focused group that understands both AI architectures and security fundamentals. This team can own AI threat modeling, policy, and incident response, and act as a partner to product and data science teams.

2. Optimize resources to secure GenAI

Inventory where GenAI is already in use (including BYOAI), identify high-risk data flows, and prioritize controls where the impact is greatest—such as customer-facing apps and workloads handling sensitive or regulated data.

3. Implement a Zero Trust strategy for AI

Apply Zero Trust principles—verify explicitly, use least privilege, and assume breach—to AI systems. That includes:

• Tight identity and access controls for AI apps and services.
• Granular permissions to reduce over-permissioned AI access.
• Continuous monitoring for anomalous use, data oversharing, and shadow AI tools.

4. Adopt a comprehensive security solution for AI

Look for integrated capabilities that address both traditional and AI-specific threats, such as:

• Data loss prevention and governance for prompts, training data, and outputs.
• Protection against prompt injection, model abuse, and insecure plug-ins.
• Visibility into AI components, supply chain dependencies, and misconfigurations.

By treating AI security as a structured transformation—rather than a bolt-on control—organizations can reimagine how they adopt GenAI, enabling innovation while keeping data protection, compliance, and trust at the center.