# Shadow AI Exposure Assessment: How Much of Your Data Is Already Leaking? > Source: https://netray.co/tools/shadow-ai-exposure-assessment > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Free Tools ## Short answer Score your organization across eight dimensions of shadow AI risk, from detection capability and policy coverage to what has already been pasted into public tools. ## Category on-prem-ai ## Tool **Kind:** assessment **Questions:** #### Do employees have a sanctioned AI tool capable enough to satisfy their daily work? **Help:** Shadow AI use is overwhelmingly a capability gap problem, not a discipline problem. People route around slow, weak, or absent official tools. **Options:** ##### Yes, a capable sanctioned tool exists and is actively promoted **Score:** 3 ##### Yes, but it is noticeably weaker than public consumer tools **Score:** 2 ##### A sanctioned tool exists but adoption is low and unmeasured **Score:** 1 ##### No sanctioned AI tool exists at all **Score:** 0 #### Can you technically detect when a corporate device submits data to a consumer AI tool? **Options:** ##### Yes, network and endpoint controls specifically cover AI chat and copilot endpoints **Score:** 3 ##### Partially, through general web monitoring not tuned for AI tools **Score:** 2 ##### We could investigate after the fact but do not monitor in real time **Score:** 1 ##### No detection capability at all **Score:** 0 #### Do you have a written, communicated policy on acceptable AI tool use? **Options:** ##### Yes, specific, current, and covered in onboarding and periodic training **Score:** 3 ##### Yes, but generic and rarely referenced after onboarding **Score:** 2 ##### A draft exists but was never formally published **Score:** 1 ##### No policy exists **Score:** 0 #### Have you audited browser extensions and desktop AI plugins installed on corporate laptops? **Options:** ##### Yes, on a recurring schedule with unapproved extensions blocked **Score:** 3 ##### Yes, a one-time audit was performed **Score:** 2 ##### We have looked informally but never systematically **Score:** 1 ##### No audit has ever been performed **Score:** 0 #### Do your DLP or CASB tools have policies specifically covering AI chat and copilot destinations? **Options:** ##### Yes, AI destinations are explicitly categorized and policy-controlled **Score:** 3 ##### Partially, general file-upload policies incidentally catch some AI traffic **Score:** 2 ##### DLP or CASB exists but was never tuned for AI destinations **Score:** 1 ##### No DLP or CASB coverage exists **Score:** 0 #### As far as you know, has proprietary code, credentials, or customer data ever been pasted into a public AI tool? **Options:** ##### No, and we have telemetry that gives us confidence in that answer **Score:** 3 ##### Not that we are aware of, but we have no way to verify it **Score:** 1 ##### Yes, at least one confirmed instance that was addressed **Score:** 1 ##### Yes, and it happens regularly with no consequence or correction **Score:** 0 #### Do departments outside IT independently procure or expense AI tools? **Options:** ##### No, all AI tool procurement runs through a single reviewed process **Score:** 3 ##### Rarely, and it gets caught during expense review **Score:** 2 ##### Occasionally, with no consistent review **Score:** 1 ##### Yes, regularly, and IT often finds out after the fact **Score:** 0 #### If a shadow AI incident happened today, would you have logs to investigate it? **Options:** ##### Yes, sufficient network and endpoint logs with adequate retention **Score:** 3 ##### Partial logs that would narrow but not fully resolve the investigation **Score:** 2 ##### Logs might exist somewhere but nobody knows exactly where **Score:** 1 ##### No, we would have essentially nothing to investigate with **Score:** 0 **Bands:** #### Wide open: shadow AI is your largest unmanaged risk **Min:** 0 **Max:** 29 **Verdict:** You have no meaningful visibility into what data employees are sending to public AI tools, no enforced policy, and no logs to investigate an incident if one is reported. Given how quickly employees adopt capable free AI tools, it is reasonable to assume proprietary data has already left the building through this path. **Recommendations:** - Stand up a sanctioned AI tool this quarter so employees have a real alternative to public tools. - Publish a short, specific, enforced AI use policy rather than relying on a general acceptable-use policy to cover it. - Turn on basic DLP or CASB categorization for known AI chat destinations immediately, even before a full rollout. - Run an anonymous survey asking what AI tools people actually use today; the honest answer will surprise you. - Centralize AI tool procurement so finance and IT see every new subscription. #### Partial coverage, real gaps remain **Min:** 30 **Max:** 54 **Verdict:** Some controls exist but coverage is inconsistent, and you likely could not confidently answer whether sensitive data has been exposed through an unsanctioned tool. The most common gap at this stage is a policy that exists on paper but is not backed by any technical enforcement. **Recommendations:** - Extend DLP or CASB policy coverage specifically to AI chat and copilot destinations, not just generic web categories. - Run a scheduled browser extension and plugin audit rather than a one-time check. - Improve the sanctioned tool so its capability gap with public alternatives closes, reducing the incentive to route around it. - Add AI-specific scenarios to security awareness training with concrete examples of what not to paste. - Confirm log retention is long enough to support an investigation that surfaces weeks after the fact. #### Managed risk, close the remaining blind spots **Min:** 55 **Max:** 79 **Verdict:** You have real policy, detection, and logging in place. The remaining exposure is usually around newly adopted tools that have not yet been categorized, and departments that procure AI tools outside the standard review process. **Recommendations:** - Review AI tool spend across expense reports quarterly to catch procurement that bypassed IT review. - Re-audit browser extensions after any major AI product launch, since adoption spikes fast. - Tighten DLP rules as new AI destinations appear; the list of consumer AI tools grows constantly. - Test your incident response process with a tabletop exercise simulating a confirmed data exposure. - Publish a lightweight fast-track review process so new AI tools get evaluated within days, not months. #### Strong shadow AI governance **Min:** 80 **Max:** 100 **Verdict:** You have a capable sanctioned tool, enforced policy, technical detection, and the logs to investigate an incident if one occurs. This is a genuinely mature posture that most organizations have not reached, and it removes one of the largest unmanaged data exfiltration paths in a modern workplace. **Recommendations:** - Keep the sanctioned tool's capability ahead of public alternatives so the incentive to go around it stays low. - Review DLP and CASB AI-destination coverage quarterly, since new tools launch constantly. - Extend the same governance model to AI features embedded inside existing SaaS products, not just standalone chat tools. - Share your governance framework across business units as the organization grows or acquires new teams. - Periodically test detection with a controlled exercise rather than assuming it still works as designed. ## Content **Intro:** This free shadow AI exposure assessment scores your organization across eight dimensions that determine whether employees are already sending proprietary data to unsanctioned AI tools, and whether you would even know if they were. It is built for IT security leads, CISOs, and compliance managers at manufacturers where engineering drawings, customer data, or export-controlled information could end up inside a consumer chatbot with a single paste. Answer eight questions covering sanctioned tool capability, technical detection, policy, and incident readiness, and get a scored risk band with prioritized fixes. Shadow AI is rarely malicious. It is an engineer under deadline pressure reaching for the fastest tool available, and that tool is usually not the one IT approved. **Sections:** #### Why shadow AI spreads faster than any other shadow IT before it **Body:** Previous generations of shadow IT required signing up for a SaaS product and usually a company email address, which left some paper trail. Consumer AI tools need none of that: a personal email, a free tier, and a paste. The capability gap is also unusually large, since frontier consumer AI tools are often more capable than whatever sanctioned internal tool IT has managed to stand up, which gives employees a genuine productivity reason to route around policy rather than a lazy one. That combination, low friction plus real capability advantage, is why shadow AI adoption curves look nothing like previous shadow IT waves. **Bullets:** - A single paste into a browser tab leaves almost no trace without AI-specific DLP categorization. - Free-tier consumer AI accounts require no corporate approval, procurement, or IT visibility to start using. - Browser extensions and desktop AI copilots often request broad permissions that quietly read page content. - Engineers under deadline pressure will choose the faster tool over the approved one unless the gap is closed. #### The eight risks this assessment measures **Body:** Each question targets a real control gap: whether a viable sanctioned alternative exists, whether you can technically detect AI tool traffic, whether policy is written and enforced rather than assumed, whether extensions and plugins are audited, whether DLP actually categorizes AI destinations, whether an exposure has already happened, whether procurement is centralized, and whether you have logs to investigate. These are the exact questions a post-incident review asks, almost always after the fact, which is precisely why answering them proactively is worth the hour it takes. #### How to read your score **Body:** A low score is common and does not mean your team is careless, it usually means AI tool governance simply has not caught up to how fast adoption happened. The fastest high-leverage fix is almost always standing up a genuinely capable sanctioned tool, because policy and detection controls only work when there is a real alternative for people to be redirected to. Detection and DLP tuning close the visibility gap; policy and training close the awareness gap; but neither survives contact with a workforce that has no acceptable tool to use instead. #### How Netray closes the shadow AI gap **Body:** Netray deploys sanctioned, on-prem AI assistants for manufacturers specifically so employees have a fast, capable, IT-approved alternative to consumer tools, one that never sends proprietary data outside your network. We pair the deployment with practical policy language, DLP tuning guidance for AI-specific destinations, and integration with your existing SyteLine, LN, or M3 data so the sanctioned tool is actually more useful than the public one, which is what makes adoption stick instead of quietly reverting to shadow use within a month. **Cta:** Get a shadow AI exposure review and a plan for a sanctioned tool employees will actually choose to use. **Faq:** #### Is shadow AI use usually malicious? **Answer:** Almost never. The overwhelming pattern is an employee trying to move faster, not trying to cause harm. Someone pastes a specification into a public AI tool to get a quick summary, or drops a chunk of code into a chatbot to debug it, without stopping to consider where that data goes afterward. The fix is closing the capability gap with a sanctioned alternative and clear policy, not punitive enforcement aimed at individuals. #### Can DLP tools actually stop data from reaching consumer AI sites? **Answer:** Modern DLP and CASB platforms can categorize known AI chat and copilot destinations and block or flag uploads to them, but the category list requires active maintenance since new AI tools launch constantly. Coverage is also weaker on personal devices and unmanaged browsers. Treat DLP as one layer among several, not a complete solution on its own; policy, training, and a viable sanctioned alternative all still matter. #### What should an acceptable AI use policy actually say? **Answer:** It should name approved tools explicitly, state clearly what categories of data can never leave the sanctioned environment such as export-controlled data, customer confidential information, and credentials, and describe the review process for requesting a new tool. Vague language about using AI responsibly does not change behavior. Specific, enforceable rules tied to real examples do, especially when paired with a sanctioned tool that is actually good enough to use. #### How do I estimate whether shadow AI exposure has already happened? **Answer:** Start with an anonymous survey asking what AI tools people actually use for work today, separate from what is officially sanctioned; the honest response rate is usually high because nobody feels individually implicated. Cross-reference with any browser extension or network traffic data you already have. If neither exists, assume exposure has occurred wherever a genuinely useful sanctioned tool has not been provided, since that gap is the single strongest predictor. ## Search terms shadow AI assessment, shadow AI risk, unsanctioned AI tools, consumer AI tool exposure, AI usage policy enforcement, DLP for AI tools, shadow IT AI risk, employee AI tool audit --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json