# Introducing shadow AI discovery: find the AI you didn't sanction > Source: > Published: 2026-07-27 07:00:00+00:00 Welcome to C1 Launch Week. This week, we're introducing the building blocks organizations need to adopt and secure AI agents. Each post covers another step: discovering the AI already in your environment, securing the credentials it carries, putting up guardrails to govern at runtime, and responding to risk once it's under management. Together, they offer a practical path to adopting AI fearlessly. We begin with discovery. **TL;DR:** Today we're launching shadow AI discovery. It helps security teams find unsanctioned AI tools, AI agents, MCP servers, and exposed credentials across the cloud and endpoints, then assign ownership and govern them. ## AI adoption outpaced governance[#](#ai-adoption-outpaced-governance) Most teams moved fast on AI to keep up. Now they can't see everything they turned on. AI didn't just add a few agents; it drove a sharp increase in the number of non-human identities in the environment, leaving security teams to catch up. You can't govern what you can't see. Much of that AI is also invisible to the tools you already run: privileged access management (PAM) guards human sessions, endpoint detection and response (EDR) watches processes, and cloud access security brokers (CASB) and data loss prevention (DLP) watch sanctioned SaaS. These tools assume an identity is a person or a sanctioned app. But the fastest-growing identities are non-human, and the least governed may be the AI agent that authenticates on its own and accumulates credentials across clouds and laptops. Without an owner, these agents fall outside existing processes. In IBM's *Cost of a Data Breach Report 2025*, one in five organizations reported a breach due to shadow AI. Among those breached through an AI model or application, 97% lacked proper AI access controls. ## Where shadow AI shows up and how C1 finds it[#](#where-shadow-ai-shows-up-and-how-c1-finds-it) Shadow AI is spread across two surfaces: endpoints, and your cloud and identity estate. The goal is the same in both places: find what's running, assign an owner, and bring it into your existing governance processes. Let's dig into what it is and exactly how C1 surfaces this information. ## Shadow AI on endpoint[#](#shadow-ai-on-endpoint) The hardest exposure to see is often on developer laptops: coding assistants people install themselves and MCP servers they run locally over `stdio` . That MCP traffic never crosses the network, so cloud scanners and proxies can't see it. ### What it finds[#](#what-it-finds) C1 scans these devices and produces a report covering three things: **The AI tools installed**, each recognized by the configuration it writes to disk.** The MCP servers each tool declares**, normalized across a dozen config dialects into one consistent picture, local`stdio` servers included, with governed, catalog-known servers told apart from shadow ones.**The credential files those tools leave behind**, recorded by type and location only. And discovery is only step one. C1 not only *finds* an exposed credential but also *secures* it: pull the secret into a governed vault and strip the plaintext from the config. But that leads into tomorrow's launch so come back to learn more. ## Shadow AI in cloud and identity providers[#](#shadow-ai-in-cloud-and-identity-providers) The second surface is the growing number of non-human identities (NHI) across your cloud and identity providers, including AI agents. ### What it finds[#](#what-it-finds-1) C1's connectors enumerate AI agents running in platforms like Salesforce Agentforce and AWS Bedrock AgentCore, alongside the broader non-human population already in your estate. That includes service principals, managed identities, app registrations, integration tokens, and service accounts across Entra, Okta, GCP, GitHub, Snowflake, Active Directory, and more. Each one is linked to an owner. Unowned, misclassified, or otherwise risky accounts are flagged, while credential health identifies expired, expiring, and unused secrets before they become standing risk. ### How C1 manages shadow AI[#](#how-c1-manages-shadow-ai) In C1, a discovered agent, MCP server, or credential becomes an access item on the same agentic control plane as employees. You can assign an owner, route it through request and approval, certify it in a review, and de-provision it when it goes stale. Governance becomes a lifecycle, not a switch. Every discovery, ownership change, and action is logged, giving you a clear answer when an auditor asks whether your AI is under control. ## This is day one[#](#this-is-day-one) Shadow AI discovery is the foundation for the rest of the week. Once every agent, tool, and credential is visible and owned, the governed path becomes the fast path. Tomorrow we secure the credentials your AI depends on, so the secrets an agent needs never sit exposed. Then we govern what your agents do at runtime. And we close the week by bringing every identity, risk, and response into one view. Four drops, one agentic control plane, the platform C1 is built to power. See what's running in your environment — book a demo at [c1.ai](http://c1.ai). Adopt AI, fearlessly. ## FAQ[#](#faq) ### What is Shadow AI?[#](#what-is-shadow-ai) Shadow AI is any AI tool, agent, or integration running in your organization without formal approval, visibility, or ownership. It includes employee-installed coding assistants, locally run MCP servers, and the non-human identities and credentials those tools create or use across cloud and endpoint environments. Because this activity often doesn't pass through traditional security controls (like PAM, EDR, CASB, or DLP), it can be hard to detect and easy to misgovern—leading to credential sprawl, untracked access paths, and increased breach risk. ### What is an MCP server (and why does it matter for security)?[#](#what-is-an-mcp-server-and-why-does-it-matter-for-security) An MCP server is a service (often local over stdio) that connects AI tools to data and systems like files, repos, and internal apps. It matters for security because it can create "invisible" access paths that don't pass through network controls, and it often relies on local tokens/config files. Without governance, MCP servers can expand an agent's effective privileges and increase credential sprawl and data exposure risk. ### What are non-human identities (NHI) and how are they related to AI agents?[#](#what-are-non-human-identities-nhi-and-how-are-they-related-to-ai-agents) Non-human identities (NHIs) are non-person accounts that can authenticate. These may be service principals, managed identities, app registrations, service accounts, and tokens. AI agents typically run using NHIs to access systems and data. As teams adopt more agents, NHIs grow quickly, and unowned or over-permissioned identities plus long-lived secrets become standing risk. ### What's coming next?[#](#whats-coming-next) Shadow AI discovery is the first of four drops this week. Next we secure the credentials your AI depends on, then govern what your agents do at runtime, and finally bring every identity, risk, and response into one view — the Agentic Control Plane. ## Sources[#](#sources) - IBM, *Cost of a Data Breach Report 2025*: one in five organizations reported a breach due to shadow AI; among organizations breached through an AI model or application, 97% lacked proper AI access controls.[https://www.ibm.com/reports/data-breach](https://www.ibm.com/reports/data-breach)