Shadow AI in 2026: Risks, Statistics, Costs & Enterprise Security

Key Takeaways

• Shadow AI is unsanctioned AI use inside an organization — and by 2026, it's no longer an edge case. It's closer to the default state of enterprise AI adoption.
• IBM's 2026 Cost of a Data Breach Report found shadow AI-linked incidents jumped from 20% to 43% of AI-related breaches year over year, with an average cost of $5.39 million per breach.
• The core risk isn't a single event — it's continuous, invisible data exposure across sensitive records, source code, and strategic information, happening one prompt at a time.
• Discovery is genuinely harder than traditional shadow IT discovery, because AI usage often hides inside sanctioned SaaS tools' background features, not just standalone apps.
• AI agents change the risk calculus. An unauthorized chatbot exposes what you type into it; an unauthorized agent can take actions — querying databases, calling APIs, sending emails — well beyond a single conversation.
• Outright bans tend to fail quietly: usage continues on personal devices and accounts, and visibility gets worse, not better.
• Governance that works in 2026 pairs monitoring and policy with genuinely usable sanctioned alternatives — the goal is "know," not just "block."
• Privacy-first data controls, applied at the point sensitive information would reach an AI system, reduce exposure regardless of which tool an employee ultimately chooses.

What Is Shadow AI?

Shadow AI is the use of AI tools, models, plugins, or embedded AI features by employees without the knowledge, review, or approval of an organization's IT or security teams. It includes everything from pasting a client contract into a free chatbot to summarize it, to installing a browser extension that quietly runs AI over whatever page is open, to using an AI coding assistant that wasn't vetted or licensed by the company.

It is the AI-era evolution of shadow IT — the older problem of employees using unapproved software or cloud services. But shadow AI carries a distinct risk profile: the tools involved don't just store or transmit data, they actively process it, generate new outputs from it, and in many cases retain it to improve a third party's model. That difference is why security and compliance teams increasingly treat shadow AI as a separate risk category rather than a subset of shadow IT.

How Big a Problem Is Shadow AI in Large Enterprises?

Direct answer: Shadow AI affects the large majority of large enterprises today, not a minority. Multiple 2026 industry surveys put unsanctioned AI usage among employees in the 60–90% range depending on how usage is defined, while only a small fraction of organizations report having formal AI usage policies or full visibility into what's actually running. The gap between adoption and governance — not the existence of AI use itself — is what defines the scale of the problem.

The pattern behind that gap is consistent across research: employee AI adoption moved far faster than enterprise governance could follow. According to Salesforce's 2026 Workforce AI Survey, 67% of employees now use AI tools at work, while only 18% of organizations report having formal AI security policies in place. PagerDuty's 2026 international Shadow AI Survey, conducted among 1,250 office professionals at companies with $500 million or more in annual revenue across Australia, Japan, the UK, and the US, found that two-thirds of office professionals had used unauthorized AI tools at work.

Three structural factors explain why the gap keeps widening rather than closing:

Consumer AI products are not built for enterprise governance. Most of the tools employees reach for were designed to ingest data and improve on it — not to support enterprise retention controls, audit trails, or data lineage documentation.

Shadow AI hides inside tools that are already approved. Many organizations have reasonably tight control at the application layer and almost no visibility into the AI functionality quietly running inside the background of SaaS products their teams already use — an embedded summarization feature, an AI-powered search bar, a "smart" autocomplete calling an external model.

Employees are solving a real problem, not creating a reckless one. When a task that used to take two hours can be done in two minutes with an unapproved tool, most employees will use it — not out of disregard for policy, but because the sanctioned alternative doesn't exist or doesn't work as well.

Shadow AI Statistics: What the Data Shows in 2026

The numbers below are drawn from named, dated studies. Where a figure comes from a survey rather than measured incident data, that distinction is noted, since the two aren't interchangeable.

Breach and incident data (IBM, 2026 Cost of a Data Breach Report): Shadow AI-linked security incidents rose from 20% to 43% of AI-related breaches year over year, based on Ponemon Institute research across 602 breached organizations in 17 industries and 16 countries between March 2025 and February 2026. The average cost of a breach involving shadow AI reached $5.39 million. Sixty-eight percent of breached organizations had no policy in place to govern AI use or manage shadow AI, and 92% of organizations that experienced an AI-related breach lacked adequate AI access controls. Separately, one in four malicious breaches in the same study were AI-enabled — deepfake impersonation, AI-generated malware, and AI-crafted phishing — a 56% increase year over year, with those attacks averaging $6 million per breach.

What this means for enterprises: these are not projections. They're measured outcomes from organizations that had already been breached, which makes the governance gap IBM documents — most breached organizations had no AI oversight policy at all — the more urgent number in the data set.

Employee adoption data (Salesforce, 2026 Workforce AI Survey): 67% of employees report using AI tools at work; only 18% of organizations report having a formal AI security policy.

Employee adoption data (PagerDuty, 2026 Shadow AI Survey): 66% of office professionals at large enterprises (revenue $500M+) report having used unauthorized AI tools at work, based on a survey of 1,250 professionals across four countries.

Healthcare-specific data (Wolters Kluwer, 2026): 40% of healthcare professionals report encountering unauthorized AI tools in the workplace, nearly 20% admit to using them, and roughly one in ten report using an unauthorized AI tool in a direct patient-care context.

Legal-specific data (Thomson Reuters, 2024): 45% of legal professionals reported using consumer AI tools for work tasks — a figure worth tracking forward given how quickly adoption has moved since, but the most recent verifiable figure available for the sector.

Data-loss pattern data (Verizon, 2026 Data Breach Investigations Report): Analysis of 858,440 DLP events involving uploads to generative AI tools found source code was the single most common data type uploaded to unauthorized AI systems, ahead of images and structured data.

What this means for enterprises: the risk isn't evenly distributed. Source code exposure concentrates in technology and engineering functions; patient-context exposure concentrates in healthcare; privileged-information exposure concentrates in legal. A generic, one-size-fits-all shadow AI policy misses where the actual exposure sits inside a given organization.

Treat any shadow AI statistic — including the ones above — as a snapshot, not a constant. Given how quickly adoption is moving, figures from even a year prior likely understate current exposure.

What Are the Biggest Shadow AI Security Risks?

Direct answer: The most significant shadow AI risks are sensitive-data leakage into third-party systems with no enterprise retention controls, loss of audit trail and accountability for AI-influenced decisions, unauthorized AI agents taking real actions rather than just generating text, and a widening gap between what regulators expect organizations to document and what shadow AI usage actually allows them to prove.

Sensitive-data leakage. When an employee pastes customer records, financial data, or unreleased product information into an unauthorized AI tool, that data typically leaves the organization's control entirely — often to a vendor with no contractual data-handling agreement with the enterprise at all.

PII and confidential business information exposure. Personal information, pricing models, M&A discussions, and internal strategy documents are all common categories that end up inside shadow AI prompts, frequently without the employee registering it as a data-handling decision at all.

Intellectual property and source-code exposure. As Verizon's 2026 DBIR data shows, source code is the single most common category of data uploaded to unauthorized AI tools — a direct threat to competitive position when that code represents proprietary logic, algorithms, or product architecture.

Credential exposure. API keys, service account credentials, and access tokens are sometimes pasted into AI tools alongside code or configuration files being debugged — turning a productivity shortcut into a potential system-access vulnerability.

Insecure or malicious AI applications. Not every AI tool employees find is a legitimate product with reasonable security practices. Some browser extensions and AI-branded apps are built specifically to harvest the data passed through them.

Prompt injection. Attackers can embed hidden instructions in content an AI tool processes — a webpage, a document, an email — that hijack the tool's behavior. Shadow AI tools, deployed without security review, typically have no defenses against this at all.

Third-party data retention. Many consumer AI products retain submitted data to improve their models by default, meaning information an employee considered a one-time query may persist indefinitely in a system the enterprise has no visibility into or control over.

Unauthorized AI agents. An agent that can take actions — not just generate text — introduces risk that scales with what it's connected to: the files it can access, the systems it can call, the workflows it can trigger.

AI supply-chain risk. Shadow AI tools often rely on their own third-party model providers and subprocessors, meaning the enterprise inherits risk from vendors it never assessed and may not even know exist.

Insecure APIs. Background AI features embedded in approved SaaS tools frequently call external APIs the security team never reviewed, creating exposure that looks, from the outside, like normal application traffic.

Compliance violations. Regulations like the EU AI Act and GDPR increasingly expect documented data lineage and AI usage governance; shadow AI usage makes that documentation impossible to produce accurately, regardless of what internal policy says on paper.

Inaccurate outputs feeding real decisions. When shadow AI tools generate wrong or fabricated information that quietly informs a business decision, the resulting risk is often invisible until the decision's consequences surface.

Lack of monitoring. Every risk above is compounded by the same root problem: without visibility into which tools are in use, security and compliance teams cannot assess exposure, let alone respond to an incident in progress.

What Does Shadow AI Cost Enterprises?

Shadow AI's cost shows up in more places than a single breach line item. It's useful to separate direct financial costs from indirect business costs, since they call for different responses.

Direct financial costs:

• Breach and incident costs. IBM's 2026 research puts the average cost of a breach involving shadow AI at $5.39 million — measurably above the overall average breach cost of $4.99 million reported in the same study.
• Regulatory fines. IBM found regulatory fines occurred in roughly one in five AI-related breach incidents, adding direct financial exposure on top of remediation costs.
• Redundant and unauthorized subscriptions. Individual employees and teams frequently pay for AI tools out of expense accounts or personal cards, creating duplicate spend the organization has no visibility into and no negotiating leverage over.
• Uncontrolled API costs. Where shadow AI usage runs through pay-per-use APIs rather than flat subscriptions, costs can scale unpredictably with usage the organization isn't tracking.
• Incident response and remediation. Investigating and containing a shadow AI-related exposure typically costs more than a routine incident, precisely because there's no existing inventory of what tool was used, what data it touched, or who else may have used it.

Indirect business costs:

• Audit and compliance overhead. Reconstructing what data went where, after the fact, is dramatically more expensive than maintaining that documentation as a byproduct of a governed AI environment from the start.
• Productivity losses from fragmented tooling. Ironically, the same organizations exposed to shadow AI risk often aren't capturing shadow AI's productivity upside efficiently either — value is scattered across unmanaged, uncoordinated tools rather than consolidated into tools the whole organization can build on.
• Vendor sprawl. Every unauthorized tool is a vendor relationship the organization has implicitly entered into without due diligence, multiplying the number of parties with some claim on enterprise data.
• Reputational and trust costs. These are the hardest to price precisely and often the most consequential — a publicized shadow AI-linked exposure of customer or patient data affects trust in ways that outlast the direct financial remediation.
• Strategic risk. Decisions made on the back of unverified or ungoverned AI outputs carry a cost that may not surface until well after the decision itself, when the underlying error becomes apparent.

None of these figures should be treated as fixed. They vary meaningfully by industry, data sensitivity, and existing governance maturity — the ranges above are directional, not a forecast for any specific organization.

Which Industries Are Most at Risk From Shadow AI?

Direct answer: Healthcare, financial services, legal, and technology sectors face the highest shadow AI exposure, because each handles data categories — protected health information, regulated financial records, privileged client communications, and proprietary source code — where unauthorized AI exposure creates not just a security incident but a specific regulatory or professional liability.

Healthcare. Patient information, clinical notes, and treatment data are subject to strict regulatory protection, and unauthorized AI use in clinical or administrative workflows creates direct compliance exposure. Wolters Kluwer's 2026 research found roughly one in ten healthcare professionals had used an unauthorized AI tool in a direct patient-care context — a use case where an inaccurate or ungoverned AI output carries consequences well beyond a typical data-handling policy violation.

Financial services and insurance. Customer financial records, transaction data, and underwriting information sit under some of the most detailed regulatory requirements of any sector. Unauthorized AI processing of this data creates exposure not just to data breach rules but to sector-specific financial regulation governing how customer data can be processed and by whom.

Legal. Attorney-client privilege and confidentiality obligations mean that pasting case details or client communications into a consumer AI tool isn't just a data-handling risk — it can jeopardize privilege itself. Thomson Reuters' 2024 research found 45% of legal professionals had used consumer AI tools for work tasks, a figure that underscores how normalized the behavior had already become before most firms had formal AI policies in place.

Technology. Source code, product roadmaps, and technical architecture represent an organization's core competitive position. Verizon's 2026 DBIR analysis of AI DLP events found source code was the leading data category uploaded to unauthorized generative AI tools — a direct line from shadow AI usage to intellectual property exposure.

Government and public sector. Citizen data, security-classified information, and public trust obligations make unauthorized AI use in government workflows a governance issue with implications beyond any single agency.

Professional services. Consulting, accounting, and advisory firms routinely handle multiple clients' confidential strategic and financial information simultaneously, multiplying the exposure surface of any single unauthorized AI interaction.

Manufacturing. Proprietary designs, supply-chain data, and operational technology information create intellectual property and, increasingly, operational-security exposure when processed through ungoverned AI tools.

Education. Student records and research data — often protected under sector-specific privacy rules — face growing exposure as AI tools spread informally through both administrative and academic use.

The core distinction: shadow IT is primarily a storage and access problem. Shadow AI is a processing, retention, and output-integrity problem layered on top of the same underlying access questions — which is why it typically demands additional controls around prompts, model behavior, and AI agent permissions that traditional shadow IT governance never had to consider.

Shadow AI vs Approved Enterprise AI

The difference between shadow AI and approved enterprise AI isn't the underlying technology — it's what surrounds it.

Unauthorized (shadow) AI typically involves:

• Unknown or unvetted vendors
• Unclear or undisclosed data retention practices
• No contractual data-processing agreement with the enterprise
• Little to no usage monitoring
• No defined accountability for outputs
• No integration with existing access controls or data classification

Approved enterprise AI typically involves:

• Vendor security and privacy assessment before deployment
• Documented data-handling and retention policies
• Defined access controls aligned to existing identity and data-governance systems
• Usage monitoring and audit logging
• Clear ownership and accountability for the AI system's outputs
• Defined, approved use cases rather than open-ended experimentation

The practical implication for enterprise leaders: the goal isn't to eliminate employee AI use — it's to close the gap between what employees are already doing and what the organization can actually govern, ideally by making the approved path the easier one.

Shadow AI and Explainability: Why Visibility Matters

Shadow AI and explainable AI (XAI) are related but distinct concepts, and it's worth being precise about the difference: shadow AI describes unauthorized usage of AI tools; explainability describes whether an AI system's reasoning can be understood and audited at all. A tool can be fully sanctioned and still lack explainability, and a shadow AI tool is, by definition, one an organization hasn't been able to assess for explainability in the first place.

The connection between the two is visibility. To evaluate whether any AI system — sanctioned or not — meets a reasonable explainability standard, an enterprise needs to be able to answer a specific set of questions:

• Which AI system was actually used?
• Who used it, and in what context?
• What data was provided to it?
• What output did it generate?
• What decision or workflow did that output influence?
• Can the interaction be audited after the fact?
• Did the interaction follow existing data-handling policy?

Shadow AI usage makes every one of these questions unanswerable by definition — there's no record to audit, because the interaction happened outside any monitored system. That's precisely why shadow AI and explainability sit next to each other on a governance agenda: an organization can't build toward explainable, accountable AI use while a meaningful share of its actual AI usage remains invisible to the systems meant to provide that accountability.

How to Discover Shadow AI Across an Enterprise

Direct answer: Effective shadow AI discovery combines technical telemetry — browser, endpoint, network, and API monitoring — with organizational signals like procurement and expense data, and it works best as a continuous process rather than a one-time audit, since new AI tools and embedded AI features appear faster than any static inventory can track.

Why is shadow AI harder to discover than traditional shadow IT? Traditional shadow IT discovery relies heavily on network traffic and SaaS account discovery — an unauthorized cloud storage account or messaging app tends to show up clearly in that data. Shadow AI is different because a meaningful share of it doesn't run as a standalone application at all. It runs as a browser extension injecting functionality into pages an employee is already authorized to visit, or as a background feature inside a SaaS tool the organization has already approved. The application layer looks normal; the AI processing happening underneath it doesn't.

Practical discovery methods enterprises are using in 2026 include:

• AI-specific application discovery tools, purpose-built to identify AI functionality rather than just SaaS account sprawl.
• Browser and endpoint telemetry, which can surface AI-related domains, extensions, and API calls that traditional network monitoring misses.
• Network and application-layer telemetry, extended specifically to look for calls to known AI model provider endpoints.
• Identity and SSO data, which can reveal AI tool sign-ups even where the tool itself sits outside traditional network visibility.
• DLP (data loss prevention) signals, tuned specifically to detect uploads to generative AI domains and endpoints — the same category of data Verizon's 2026 DBIR analysis drew on.
• API monitoring, to catch AI functionality embedded inside otherwise-approved software.
• Procurement and expense data review, to catch AI subscriptions purchased outside formal IT procurement.
• Anonymous employee surveys, which frequently surface tools that technical discovery alone misses, particularly where usage happens on personal devices.
• A living AI application inventory, mapping every discovered tool against approved-vs-unapproved status rather than treating discovery as a single project with an end date.
• Continuous monitoring, since the tools available to employees — and the AI features embedded inside already-approved software — change constantly.

How Can Enterprises Reduce Shadow AI Risk?

The organizations managing shadow AI well in 2026 have largely made the same shift: from "block" to "illuminate." A ban without a usable alternative doesn't eliminate the behavior — it just makes it harder to see, since employees who need a tool badly enough will typically find one that doesn't trip the current firewall and stop mentioning it.

Practical steps that work in combination:

• Establish an AI acceptable-use policy that's specific enough to actually guide behavior — vague guidance rarely changes what people do.
• Provide approved AI tools that genuinely meet the need employees are currently solving with unauthorized ones. If staff are using a public chatbot, an enterprise-grade equivalent with real data controls closes the gap far more effectively than a policy memo.
• Maintain a living AI application inventory, built from the discovery methods above and reviewed on a recurring basis, not a one-time project.
• Classify sensitive information so employees and systems alike know what can and can't be shared with any AI tool, sanctioned or not.
• Control what sensitive data can reach external AI systems, ideally through technical enforcement rather than policy alone.
• Assess AI vendors before approval, covering data retention, subprocessors, security practices, and contractual terms.
• Monitor AI application usage continuously, treating it as an ongoing security and governance function rather than a periodic audit.
• Govern AI agents specifically, with defined permissions, identity, and human-approval thresholds for autonomous actions.
• Train employees on what's actually at risk, since most shadow AI usage comes from a gap in awareness rather than intent to cause harm.
• Establish an incident response process specific to AI, since a generic breach playbook often doesn't account for how AI tools retain and process data differently than traditional software.
• Review the AI landscape continuously, since new tools, new embedded features, and new agentic capabilities appear faster than any static governance program can anticipate.

Shadow AI Agents: The Next Enterprise Risk

Most shadow AI discussion to date has focused on a fairly contained interaction: an employee pastes information into a chatbot and receives text back. Shadow AI agents represent a meaningfully different category of risk, because an agent doesn't just respond — it acts.

An unauthorized AI agent, depending on what it's connected to, can potentially:

• Access files and internal document repositories
• Query databases directly
• Call internal or third-party APIs
• Send emails or messages on a user's behalf
• Execute multi-step workflows without further human input
• Interact with other business applications
• Make and act on recommendations, rather than simply generating them
• Take real, consequential actions inside enterprise systems

Gartner projects that 40% of enterprise applications will feature task-specific AI agents by the end of 2026, up from under 5% in 2025 — a pace of adoption that outstrips the governance most organizations have built for standalone chatbot use, let alone for autonomous agents with system access.

The governance questions this raises are fundamentally about permissions and identity rather than content: What can this agent access, and is that access scoped to only what its function requires? Which actions can it take autonomously, and which require a human checkpoint first? Is the agent's activity logged in a way that supports audit after the fact? Who is accountable when an agent takes an unintended action? An organization that has never inventoried its standalone shadow AI usage is, by definition, even further behind on answering these questions for shadow AI agents — because the consequences of an ungoverned agent compound faster than the consequences of an ungoverned chatbot.

Shadow AI, Data Privacy and Compliance

Shadow AI usage intersects with data privacy and compliance obligations in ways that depend heavily on the specific data involved and the regulatory context — the language below is intentionally cautious, since applicability varies by jurisdiction, sector, and processing activity.

Where personal data is entered into an unauthorized AI tool, that may constitute a data processing activity the organization did not authorize, document, or have a legal basis for under frameworks like the GDPR — particularly where the tool retains that data or uses it for further model training without consent. Data minimization principles, which generally call for limiting personal data processing to what's necessary for a defined purpose, are difficult to demonstrate compliance with when the actual scope of AI-driven data processing across an organization isn't known.

Vendor risk compounds this: an unauthorized AI tool is, in data protection terms, an unassessed third-party processor. Depending on the processing context, that can raise questions about data transfer mechanisms, subprocessor disclosure, and retention limits that the organization has had no opportunity to review.

Frameworks like the EU AI Act increasingly expect organizations to document data lineage — where data went, what processed it, and under what conditions — for AI systems used in relevant contexts. Shadow AI usage, by its nature, makes that documentation incomplete regardless of how thorough an organization's written policies are, because the actual usage sits outside the systems generating that documentation.

None of this amounts to a claim that shadow AI use automatically constitutes a violation of any specific law — that depends on the data involved, the jurisdiction, the processing purpose, and facts specific to each situation. What it does mean is that shadow AI meaningfully increases the difficulty of demonstrating compliance, which is often what regulators and auditors actually test for.

Shadow AI Governance Maturity Model

Enterprises tend to move through five recognizable stages as they build out shadow AI governance:

Level 1 — Unaware. The organization has no formal visibility into AI usage and no policy addressing it. Shadow AI usage is happening, but leadership has not yet acknowledged its scale.

Level 2 — Discovery. The organization begins actively identifying AI tools in use through technical discovery, surveys, and procurement review, building an initial inventory without yet having controls in place to act on it.

Level 3 — Policy. A formal AI acceptable-use policy exists, defining what data can and cannot be shared with AI tools and which tools are approved — but enforcement and monitoring remain limited.

Level 4 — Control. Technical controls are in place: access management, data classification enforcement, vendor assessment requirements, and monitoring that can detect policy violations as they happen rather than after the fact.

Level 5 — Continuous Governance. AI governance is integrated into ongoing operations rather than treated as a project — usage is monitored continuously, agent permissions are actively managed, vendor risk is reassessed on a schedule, and the organization can produce accurate documentation of AI-driven data processing on demand.

Before advancing a stage, an organization should be able to demonstrate the outcomes of the current one — an inventory that's actually current, a policy that's actually enforced, controls that actually generate audit-ready logs — rather than treating each stage as a box to check once and move past.

How Privacy-First AI Controls Can Reduce Shadow AI Data Exposure

Discovery and policy solve the visibility problem. They don't, by themselves, solve the exposure problem — because even sanctioned AI use still involves sensitive data reaching AI systems, and even a well-governed AI environment benefits from reducing what's exposed in the first place.

This is where privacy-first technical controls play a distinct role: applying data protection — anonymization, redaction, or tokenization of sensitive information — at the point content would reach an AI system, before it becomes a retained prompt sitting inside a third-party model provider's infrastructure. This doesn't eliminate the governance work described above; an organization still needs discovery, policy, vendor assessment, and monitoring. What privacy-first controls change is the consequence when sensitive data does reach an AI system, sanctioned or not — the data reaching the model has already had its most sensitive elements protected, rather than exposed in the clear.

How Questa AI Fits Into Privacy-First Enterprise AI

Questa AI approaches enterprise AI with privacy and data protection built into the workflow rather than added on afterward. Its focus is on secure data processing, sensitive-data anonymization, and privacy controls designed to reduce the exposure of sensitive information when that information reaches AI systems — the technical layer that supports, rather than replaces, the broader governance work described throughout this article.

That distinction matters. Questa AI doesn't discover shadow AI usage across an organization, and it isn't a substitute for an AI acceptable-use policy, vendor assessment process, or agent governance program. What privacy-first infrastructure like this can do is reduce what's actually at stake when sensitive data does reach an AI system — which is a meaningful part of the exposure equation, even inside a well-governed enterprise AI environment, and a genuinely useful complement to the discovery and policy work an enterprise needs to do regardless of which technical controls it adopts.

Frequently Asked Questions

Conclusion

Shadow AI in 2026 isn't a fringe security concern anymore — it's a measurable driver of breach frequency and cost, documented in IBM's own breach research, and a governance gap present at most large enterprises regardless of sector. The organizations managing it well have stopped trying to eliminate employee AI use and started building visibility into it: discovering what's actually running, understanding where the highest-risk data categories sit, and replacing prohibition with sanctioned alternatives employees will actually choose.

What ties all of this together — discovery, policy, agent governance, and compliance — is data. Every shadow AI risk ultimately comes back to what sensitive information reaches an AI system and what happens to it once it does. Privacy-first technical controls won't discover an organization's shadow AI usage or write its acceptable-use policy, but they do address the part of the exposure that persists even after governance is in place: reducing what's actually at stake in the moment sensitive data meets an AI system. That's the layer Questa AI is built to support — not a replacement for shadow AI governance, but a meaningful part of making it work.