Why Employees Turn to Shadow AI, and How to Bring Them Back
A growing concern as AI moves into the workplace is distrust over how employees are using it. People are calling the phenomenon "shadow AI": AI tools used for work without approval from IT, security or finance.
The main issue is data security. An employee who handles sensitive customer data, financial information or trade secrets could upload a sensitive document to their personal AI account; unless the employee has turned the setting off, the provider can use that info to train future models.
Many companies have started trying to combat shadow AI by reining in which models employees can use. They subscribe to one enterprise plan, and everyone is expected to abide by that plan's controls and token limits.
So far, the controls aren’t working very well. Two out of three office workers have used an AI tool at work that they believed their company had banned, according to the PagerDuty Shadow AI Survey, which Wakefield Research conducted among 1,250 office professionals in the US, UK, Australia and Japan. In the same survey, 77% said their company's AI restrictions limit their career growth.
Your business isn't powerless, though. The fix is to make the approved route easier than the workaround: give employees access to the models they need, be clear about which tools are off-limits and put AI spending somewhere finance can track it. In this guide, we’re going over how efficacious AI control sometimes means giving your team more freedom, and what your IT and finance departments can do to feel more comfortable granting the freedom needed to dissuade shadow AI.
Why employees go off-script with AI usage
Shadow AI can refer to two different kinds of use: tools the company hasn't approved at all, or personal accounts for tools the company may already license. Whether the decision to go off-policy is preference or circumstance, it happens regardless, and it can put customer data at risk in the process. Verizon's 2026 Data Breach Investigations Report found 45% of employees now use AI regularly on corporate devices, and two-thirds of those users reached AI services through personal accounts rather than corporate ones.
The impetus to use shadow AI is twofold. First, consumer AI plans are cheap for the high-powered work they can do. At about $20 a month, a personal subscription costs little enough that many employees would rather pay for it themselves than file an expense report or wait for access to the company plan.
Second, a personal account doesn't feel like a work tool. Plenty of people treat Claude as the work tool and ChatGPT as the personal assistant (73% of consumer ChatGPT messages are unrelated to work, according to OpenAI's 2025 usage study). But dropping a company file into a personal ChatGPT account instead of the company's Claude Enterprise account is a matter of data security, not just preference.
Finance and security teams need to know whether employees are using the business account the company pays for. Spending data is one of the few ways to find out.
Zylo, which sells SaaS management software, found in its 2026 SaaS Management Index that company spending on AI-native apps rose 108% in a year, while expense-based software spending rose 267%. ChatGPT became the most expensed app in Zylo's data, and that doesn't count anyone using a personal plan for a little after-hours number crunching.
An employee's own AI purchase reaches the company in one of three ways: on a manager's corporate card, through an expense report or on a personal card that never gets filed. The first two put the charge on the company's books, often miscoded as general software. The third leaves the company with no record of it at all.
The most reliable way to know whether employees can get to the business plan, and are using it, is to give them direct access with clear rules: an approved tool, a company card to pay for it, and a clear policy.
How to get shadow AI under control
Step 1: Make an AI usage policy
If you don't have one already, make one. You wouldn't be alone: only 26% of U.S. employees say their organization has communicated a clear plan for using AI, according to Gallup. We wrote about AI governance frameworks and their implications for finance recently; read it here. The takeaway from that piece applies directly here: the rules you make are only as good as the tools you have to enforce them.
Step 2: Determine which AI models your team will use
This is perhaps the most complicated step in the entire process. Choosing a model has a lot of factors to consider: which output do you prefer, whose pricing seems fairest, which platform works best with your business's existing tools? If your company runs on Excel spreadsheets and everyone in the building is on a Windows desktop, maybe Copilot is the right choice. If everyone is using API connectors and the team should have some coding independence, Claude could be a better option.
You can also switch between several. Microsoft 365 Copilot's model selector lets users choose between OpenAI and Anthropic models under the same license and data protections, and teams working through APIs can use a model router like OpenRouter to send requests to different models.
Step 3: Make it easier for your employees to access and manage that AI
There should be a testing phase for a bit. It isn't about making sure your employees know how to talk to AI correctly, although that can be helpful, but whether there are misconceptions about how and when AI should be used. Are people having a hard time understanding skills and plugins? Are people pushing code changes they shouldn't? Are they switching between models when no one is looking because they don't like the model you chose? Each of these observations is a cause for action during the pilot, because catching them now can prevent security issues, overspending and shadow AI use down the line.
Step 4: Give everyone an AI company card
You can't control what you can't see. If everyone at work can put their AI subscriptions on a dedicated company card, you're on your way to having the visibility you need to keep control. With Slash, you can spin up unlimited virtual cards, set each one's spending limit within a couple cents of the employee's approved AI spend, and then you don't have to worry about it.¹ If someone asks for access to a new model, you can create a new card for that subscription and name it "John - Claude" to keep track.
Step 5: Monitor the situation
If employees regularly complain about their AI access and the model the company uses, it can be worth testing access to multiple models. Freedom begets compliance: the more ability your team has to use the tools they want or need, the less they'll feel inclined to go behind your back to get them. They're likely going to use the tool they want anyway. It's up to you to decide whether you'll enable it or let it happen out of your sight.
The standard in finance
Slash goes above with better controls, better rewards, and better support for your business.
What not to do: ban AI tools outright
Some companies respond by banning AI tools outright: blocking the domains, tightening the policy and disciplining employees who break it. PagerDuty's survey suggests enforcement is already common. Among respondents who used AI tools they believed were prohibited, 53% were told informally to stop and 48% faced formal consequences such as a warning.
A ban takes away the company's way of paying for AI tools, but it doesn't change why employees use them. Redress Compliance, a software licensing advisory firm, argues in its Shadow AI Spend Report 2026 that the blanket bans it audited pushed AI purchases off corporate cards and onto personal accounts, which removed finance's main record of AI spending.
As mentioned in step 2 above, consider implementing a router system for different AI models. They can cut down on costs, because they can automatically route to the most efficient model for a task based on the prompt. They can make it so your employees have more access to the model they prefer, so no one is as inclined to ste up a secret workstation. And last, if everyone is using the company models more consistently, your spend is better trascked and your data is more secure.
What finance can track once AI spend is on company cards
Once AI purchases go through company cards, finance can manage them the way it manages other spending:
Budget accuracy
AI pricing is generally usage-based, which can make it difficult to forecast – this year, multiple companies have disclosed runaway AI spend that went well beyond expectations. In October, Vegalabs, a two-person software company in Norway, published an account of a roughly $17,600 Claude bill it believed its Microsoft startup credits would cover. They didn't, and the charges went to the company's credit card instead. A card dedicated to that tool, with a monthly limit, would have stopped the charges at the limit.
Audit records
Auditors, investors and acquirers expect to see where money went and why. A ledger full of small, miscoded AI charges, plus an unknown amount paid from personal accounts, makes software spend hard to defend in a board meeting or a due diligence review.
Duplicate spending
When employees expense their own subscriptions, a company can end up paying for the same tool many times over, sometimes alongside an enterprise license it already holds. Card data sorted by merchant shows where that's happening, so finance can move those accounts onto one plan.
Breach costs
IBM's 2026 Cost of a Data Breach Report found shadow AI was involved in 43% of the security incidents it studied, up from 20% a year earlier. Breaches involving shadow AI cost an average of $5.39 million, compared with $4.99 million across all breaches. Moving employees from personal accounts to business plans puts company data under terms the company negotiated, and the company pays for a breach whether or not it approved the tool involved.
How Slash helps keep AI spending in one place
Ask a few questions about your company:
- When an employee wants access to a different AI model, how many one-off decisions need to be made?
- Is there an AI usage policy they can check before asking IT?
- Is their existing AI spend already on a corporate card in their name, properly categorized, so finance knows how much that employee spends on AI?
- And if the answer to their request is no, will they go behind your back and put customer data at risk?
Slash can help your business get better answers to these questions.
With Slash, each AI subscription can get its own corporate card, capped just above the plan price and named for the employee and the tool. The card list doubles as a record of who uses which AI tools and what each one costs. If a vendor raises its price or an agent starts running up usage charges, the card stops at its limit, and the overage never reaches an employee's personal statement.
Card groups let finance set spending rules by team, so engineering's coding tools and marketing's writing tools each stay inside their own budgets. Multi-step approvals send requests for new tools to the right reviewer, and real-time alerts flag charges that don't match a card's usual pattern. With custom accounting mappings you set up once and reuse, AI spending gets tracked as its own category instead of blending into a general ‘software’ bucket.
Slash Pro costs a flat $25 a month with no per-user fees, so giving every employee a card for AI doesn't raise the platform bill, and Slash cards earn up to 2% cash back on eligible spend.
Employees are going to keep trying new AI tools. With a card ready for each one, your business can curb shadow AI. Get started with Slash below and put every AI tool your team uses on a company card with its own limit today.
Apply in less than 10 minutes today
Join the 10,000+ businesses already using Slash.
