Full Report
As organizations rush to deploy AI, enterprise defenses are struggling to keep up. This blog explores the emerging AI exposure gap — the widening divide between innovation and protection — and what security leaders can do to close it.Key takeaways:The AI exposure gap is widening as most organizations adopt AI faster than they can secure it, creating a gap between innovation and security. Many organizations are aligning with frameworks like the EU AI Act and NIST AI RMF and adopting best practices, but few are translating those efforts into real protection — with only a small share encrypting AI data, testing AI systems, or enforcing identity controls. Tenable’s latest Insight Brief, AI Adoption Outpaces Security: The AI Exposure Gap, explains how proactive defense and continuous visibility can help organizations secure AI pipelines and reduce exposure.AI isn’t just reshaping how organizations innovate, it is rewriting the rules of risk.As businesses rush to adopt AI tools and integrate them into operations, the pace of innovation has outstripped the pace of protection. The result? A growing AI exposure gap that’s leaving critical systems and data vulnerable.AI’s rapid integration into everyday business tools like productivity apps, browsers, and cloud services creates invisible pathways for attack. As AI becomes embedded into enterprise ecosystems, these same connection points between systems, models, and data expand the attack surface, exposing new risks beyond the models themselves.This blog takes a closer look at findings from Tenable’s new Insight Brief, AI Adoption Outpaces Security: The AI Exposure Gap, which builds on The State of Cloud and AI Security 2025 report, developed in collaboration with the Cloud Security Alliance (CSA). While the report mapped the broad landscape of cloud and AI security, this brief dives deeper into one urgent theme: the widening divide between AI innovation and enterprise readiness — and what security leaders can do about it.Nearly nine in ten organizations (89%) have adopted AI in some form. More than half (55%) of organizations are running AI in production, and another 34% are in pilot phases. Yet, one in three (34%) have already suffered an AI-related breach.Two of the three top causes of these breaches — exploited vulnerabilities (21%) and insider threats (18%) — are hardly unique to AI, a reminder of the importance of proactive exposure management and cybersecurity best practices.Top Causes of AI Workload Breaches: Exploited Vulnerabilities and Security Flaws Source: State of Cloud and AI Security 2025, commissioned by Tenable and developed in collaboration with the Cloud Security Alliance. Each new AI model, dataset, and integration expands the attack surface, particularly across hybrid and multi-cloud environments where visibility is fragmented and risk assessments lag behind adoption. The lesson: AI exposure is now a measurable enterprise risk — amplified by weak identity controls and limited visibility across hybrid and multi-cloud environments.Compliance ≠ securityOrganizations are making meaningful progress by aligning with frameworks like the EU AI Act and NIST AI RMF. While 51% follow these frameworks, only 22% encrypt AI data and 26% conduct AI-specific security testing, such as red-teaming.The result: Organizations may check the box on compliance while leaving critical data and AI pipelines vulnerable. Compliance and security often overlap, but they serve different goals. Source: State of Cloud and AI Security 2025, commissioned by Tenable and developed in collaboration with the Cloud Security Alliance. Other steps organizations are taking, beyond compliance, include adopting industry best practices, conducting regular audits of AI model integration protections, and implementing AI-specific identity and access controls, all of which play a role in reducing risk.How to close the AI exposure gapTo bridge the gap between AI innovation and security, organizations should:Treat AI as a dynamic attack surface and continuously monitor for anomalous activity.Strengthen foundational controls like encryption, access management, and model integrity validation.Integrate AI exposures into unified, proactive risk management strategies across cloud and hybrid environments.Security teams need unified visibility to manage this new layer of risk effectively, moving beyond compliance checklists to real-world resilience.How Tenable can helpTenable provides unified exposure management that brings together cloud, identity and AI risk insights into a single view. Tenable AI Exposure, available in the Tenable One Exposure Management Platform, gives you visibility into how your teams use AI platforms and where that usage could put your data, users, and defenses at risk. Together with Tenable AI Aware, which uncovers AI tools across your environment, Tenable now provides one of the first end-to-end solutions to both discover and secure AI platform usage as part of your exposure management program.Learn moreDownload The State of Cloud and AI Security 2025View the Insight Brief: AI Adoption Outpaces Security: The AI Exposure Gap
Analysis Summary
# Best Practices: Closing the AI Exposure Gap
## Overview
These practices address the widening "AI exposure gap"—the divide where AI innovation outpaces an organization's ability to secure it. The goal is to move beyond mere compliance with frameworks like the EU AI Act or NIST AI RMF to achieve real-world resilience by strengthening foundational controls, increasing visibility, and managing AI risk proactively as a dynamic attack surface.
## Key Recommendations
### Immediate Actions (Focusing on Foundational Controls & Visibility)
1. **Establish AI Usage Visibility:** Deploy tools (like Tenable AI Aware) to actively discover and inventory all AI tools, platforms, and endpoints currently in use across the enterprise, especially in hybrid/multi-cloud environments.
2. **Strengthen Foundational Controls:** Immediately verify and strengthen core security mechanisms specific to AI workloads, focusing on:
* Implementing or verifying **Data Encryption** for sensitive AI datasets (only 22% of organizations are reported to be doing this).
* Enforcing rigorous **Identity and Access Management (IAM)** controls for all users and systems interacting with AI models and data.
3. **Treat AI as an Attack Surface:** Begin treating all AI models, datasets, and integrations as a dynamic extension of the overall IT attack surface that requires continuous monitoring.
### Short-term Improvements (1-3 months)
1. **Mandate AI-Specific Security Testing:** Schedule and conduct AI-specific security testing, such as red-teaming, for models currently in production or pilot phases (moving past the 26% adoption rate mentioned in the context).
2. **Integrate AI Risk into Unified Management:** Integrate discovered AI exposures, platform usage data, and related risks into existing unified **Exposure Management** strategies that cover cloud, identity, and traditional IT assets.
3. **Control Model Integrity:** Implement validated processes for auditing the integrity of integrated AI models to prevent supply chain risks associated with the models themselves.
### Long-term Strategy (3+ months)
1. **Develop Continuous Monitoring for Anomalies:** Establish ongoing telemetry and anomaly detection systems specifically designed for AI pipelines and workloads to ensure early detection of adversarial activity or configuration drift.
2. **Achieve Risk-Based Prioritization:** Mature the risk management program to prioritize AI-related exposures based on potential impact and exploitability, rather than solely following compliance checklists.
3. **Adopt Integrated Exposure Management:** Fully implement a unified exposure management platform that correlates data from AI security tools, cloud security posture management (CSPM), and identity governance to provide holistic cyber risk context.
## Implementation Guidance
### For Small Organizations
- Prioritize discovery of basic AI usage (SaaS tools, developer environments).
- Focus immediate efforts on ensuring strong, MFA-enabled identity controls for any access to AI services or data storage.
- Align with the simplest necessary security requirements dictated by AI framework adoption (e.g., mapping NIST AI RMF basic requirements to existing policies).
### For Medium Organizations
- Implement a centralized system to track and govern AI model deployment authorization.
- Conduct initial AI red-teaming exercises on production or high-value pilot models.
- Start breaking down fragmented visibility by integrating Cloud Security data with preliminary AI usage logs.
### For Large Enterprises
- Deploy end-to-end exposure management that unifies visibility across complex hybrid/multi-cloud AI deployments.
- Establish formal governance bodies responsible for AI risk acceptance and control enforcement.
- Automate the validation of model integrity and data lineage as part of Continuous Integration/Continuous Deployment (CI/CD) pipelines.
## Configuration Examples
*Note: The article focuses on strategic adoption gaps, not specific technical command configurations. The following are required controls derived from the recommendations:*
| Control Area | Action Required |
| :--- | :--- |
| **Data Encryption** | Verify that datasets used for training and inference are encrypted both at rest (storage/databases) and in transit (API calls). |
| **Identity Controls** | Enforce Multi-Factor Authentication (MFA) for all administrative and developer access to AI platforms and associated cloud accounts. |
| **Model Integrity Validation** | Establish automated integrity checks (e.g., cryptographic hashing verification) on deployed model artifacts before runtime execution. |
| **Monitoring** | Configure active alerts in security information and event management (SIEM) systems for unusual data access patterns originating from or terminating at AI workloads. |
## Compliance Alignment
The recommendations align with—and aim to supersede the insufficiency of—the following frameworks:
* **NIST AI RMF (Artificial Intelligence Risk Management Framework):** By emphasizing proactive defense, continuous monitoring, and treating AI as a dynamic surface, security efforts move beyond the RMF documentation phase into active risk management functions (Govern, Map, Measure, Manage).
* **EU AI Act:** Compliance efforts should be augmented by these steps to meet the underlying safety and robustness requirements implied by the Act, specifically regarding data governance and security testing.
## Common Pitfalls to Avoid
1. **Confusing Compliance with Security:** Do not assume that aligning with framework documentation (e.g., 51% adherence) equates to actual protection. Prioritize implementing controls like encryption and testing over simply documenting alignment.
2. **Ignoring Non-Model Risks:** Do not focus only on the model itself. The expansion of the attack surface happens across **integrations, data pipelines, and the underlying cloud/hybrid infrastructure.**
3. **Fragmented Visibility:** Avoid managing traditional vulnerabilities, cloud security, and identity separately from AI exposures. This leads to blind spots where rapid AI deployment masks emerging risk.
4. **Treating AI as Static:** Do not baseline security controls and then walk away. AI exposure is dynamic due to continuous model updates, new data ingestion, and evolving integrations.
## Resources
- **Frameworks for Guidance:** NIST AI RMF and EU AI Act (for regulatory baseline).
- **Exposure Management Platforms:** Solutions offering unified visibility across Cloud, Identity, and AI Risks (e.g., Tenable One Exposure Management Platform).
- **Research Material:** Tenable Insight Brief: *AI Adoption Outpaces Security: The AI Exposure Gap*.
- **Discovery Tooling:** Tools capable of uncovering deployed AI tools across the environment (like Tenable AI Aware).