86% of organizations installed third-party code packages with critical-severity vulnerabilities. The Cloud and AI Security Risk Report 2026 shows that 70% of organizations integrated at least one AI or Model Context Protocol third-party package, embedding AI deeply into applications and infrastructure, often without centralized security oversight or visibility. This adoption creates slop-code risks because developers frequently trust and integrate AI-generated or community-sourced code with minimal technical scrutiny. 13% of organizations deployed packages with a known history of compromise, such as the s1ngularity or Shai-Hulud worms. Engineering speed now exceeds the ability of humans to assess and prioritize risks.
Developers use generative AI to write code, a process called vibe coding. This creates risks because developers often trust AI-written code more than code they write themselves. 13% of organizations have deployed packages with a known history of compromise. The industry faces a complexity gap because autonomous AI agents, ephemeral cloud workloads, and complex supply chains interact at machine speed. Threat actors use AI to accelerate reconnaissance, automate actions, and scale workflows.
Identity risks and the machine perimeter
The security perimeter dissolved as machine-to-machine interactions grew. The attack surface shifted from human users to a 100-to-1 ratio of machine and non-human identity counts. 82% of enterprises have unknown AI agents running in their IT infrastructure, and 65% of organizations experienced AI agent-related incidents in the past 12 months. These incidents caused data exposure for 61% of respondents and operational disruption for 43%. 41% of respondents said they discovered previously unknown agents multiple times in the past year. Non-human identities, such as AI agents and service accounts, pose a risk at 52% compared with 37% for human users. This creates dangerous combinations of permissions and access that fragmented security tools often fail to detect.
The rise of agentic AI, which includes autonomous entities that perform tasks and execute code with administrative-level privileges, drives this explosion. If an agent is overprivileged, an attacker can use it to exfiltrate data at machine speed without ever compromising a human credential. 65% of organizations possess unused or unrotated cloud credentials known as ghost secrets, and 17% of those carry critical administrative privileges. 49% of identities holding critical-severity excessive permissions are dormant, increasing the risk of unnoticed exploitation. To protect the orchestration brain, organizations must secure Infrastructure as Code (IaC) state files. If a system defines an AI API key or a database credential without marking it as sensitive, these files store them in plain text. 29% of high-risk workload combinations consist of workloads that are publicly accessible, critically vulnerable, and highly privileged. 92% of executives report business-impacting compromises. Regarding AI agent governance, 38% of respondents said actions exceeding scope require human approval and 24% require it be logged. Only 11% will automatically block the action. How can organizations secure the machine perimeter?
Market competition and Tenable’s response
Wiz, which Google acquired for 32 billion dollars in March 2026, holds a 0.02% market share in the cloud-security category. Most Wiz customers work in companies with 10,000 or more employees. Tenable competes with Wiz and Orca Security through its Tenable One platform. Tenable AI Exposure provides visibility into how teams use AI platforms like ChatGPT Enterprise and Microsoft Copilot. This tool is agentless and works in minutes. You should probably verify if your current tools see the same assets.
| Asset Type | Asset Class | Description |
|---|---|---|
| Users | Account | User identity interacting with AI applications like ChatGPT |
| Agents | AI Agent | Automated bots or scripts configured to perform tasks |
Tenable AI Exposure scans for two primary asset types: Users and Agents. Tenable One Foundation provides visibility across AI security and vulnerability management, while Tenable One Advanced adds business context and attack path analysis. Tenable claims its plugin library exceeds 319,000, though independent analysis suggests a count of 219,000. 80% of cloud breaches in 2025 resulted from basic mistakes like misconfigurations or exposed secrets.
