AI is creating two identity problems at once. Inside the organization, employees use AI tools nobody approved, and sensitive data flows into systems the security team cannot see. Outside, criminals use the same technology to manufacture people who do not exist and to write phishing lures that no longer look like phishing. Each problem makes the other worse.
Shadow AI leaks the details that make impersonation convincing. Synthetic identities and AI-written phishing then turn those details into account takeovers, fraudulent onboarding and payment fraud. Treating them as separate projects, one owned by IT governance and the other by fraud or security, leaves a gap between them. This article explains how the two connect and what a joined-up response looks like.
Shadow AI: the leak you did not approve
Shadow AI is any AI tool, model or integration used for work without the organization’s knowledge or approval: a personal chatbot account used to summarize a contract, a browser extension that reads every page an employee opens, a developer’s unsanctioned API key wired into an internal app. The motive is almost always productivity, not malice, which is why it spreads so quickly.
The risk is measurable. IBM’s 2025 Cost of a Data Breach Report, published in July 2025, found that one in five organizations studied had suffered a breach involving shadow AI, and that organizations with high levels of shadow AI paid roughly $670,000 more per breach than those with little or none. Among organizations that reported breaches of AI models or applications, 97 percent lacked proper AI access controls, and 63 percent of breached organizations had no AI governance policy or were still writing one.
The best-known early warning came in 2023, when Samsung restricted generative AI tools after engineers pasted confidential source code into a public chatbot. The pattern has since repeated in quieter ways: customer lists, HR records and internal org charts pasted into tools whose data retention terms nobody read. We cover the day-to-day version of this in Shadow AI in 2026: What Your Team Is Pasting Into ChatGPT.
Synthetic identities and AI-written phishing
A synthetic identity combines real fragments, such as a stolen Social Security number or a real address, with invented details and an AI-generated face. It is built to pass remote verification: a generated ID image, a selfie that matches it, a plausible credit history grown slowly over months. Criminals use these identities to open accounts, obtain credit, launder money, and increasingly to apply for remote jobs or register as vendors.
Phishing has improved in parallel. Language models write fluent, personalized messages in any language and can mimic a specific executive’s tone if given samples. IBM’s 2025 report found that 16 percent of breaches involved attackers using AI, most often for phishing or deepfake impersonation. The spelling mistakes and awkward phrasing that awareness training taught people to spot are disappearing.
How the two threats feed each other
The connection is data. Convincing impersonation needs context: who approves payments, which vendors you use, how your CFO signs emails, what projects are underway. Much of that sits in exactly the documents employees paste into unsanctioned AI tools. Once it leaves your control, you cannot know where it is stored, who can access it, or whether it ends up in a breach of the AI provider or a compromised browser extension.
In the other direction, synthetic and impersonated identities are how attackers get inside to find more data. A fake vendor gets onboarded and receives access to a supplier portal. A fraudulent remote hire gets a laptop and a login. A phished employee’s session token opens their AI assistant’s chat history. Every new identity that slips through widens the pool of information available for the next attack.
That loop is why traditional defenses struggle. Perimeter controls do not see data leaving through a browser to a legitimate AI service, and identity checks built on “does the photo match the ID” do not catch a face that was generated to match.
A joined-up defense
The goal is to control what data reaches AI tools and to make identity decisions depend on more than a single document or face. Practical steps:
- Offer an approved AI option. Banning AI outright pushes use underground. Provide a sanctioned tool with business data protections so employees have a safe default.
- Discover what is already in use. Review browser extensions, OAuth grants, SaaS sign-ups and network traffic to AI services to build an inventory of shadow AI before writing policy around it.
- Write a short AI use policy and classify data. Spell out which data categories may never go into external AI tools, and back it with data loss prevention rules that flag or block sensitive uploads.
- Strengthen identity proofing. For customer, vendor and employee onboarding, combine document checks with liveness detection, data consistency checks and behavioral signals such as device, typing and navigation patterns that are hard for synthetic identities to fake consistently.
- Verify out of band for high-risk requests. Payment changes, credential resets and data requests get confirmed through a separate, known channel, whatever the request looks like.
- Retrain staff for AI-era lures. Teach employees that a flawless email or familiar voice proves nothing, and that the verification procedure applies to everyone.
The trade-off is between speed and control. Employees adopt AI because it saves time, and heavy restrictions will be worked around. The organizations that manage this well make the safe path the easy one. Our AI consulting services help teams roll out approved AI tools with governance built in, and our compliance services cover identity proofing and data protection controls. For the deepfake side of identity fraud specifically, see Deepfakes Are on the Rise.
Frequently asked questions
Is shadow AI the same as shadow IT?
It is a subset with a sharper edge. Shadow IT covers any unapproved software. Shadow AI is riskier because users actively feed it sensitive content, and some tools retain or learn from what they receive.
Who is targeted by synthetic identity fraud?
Banks and lenders are the most common targets, but any business that onboards customers, vendors or remote employees without meeting them in person is exposed, including staffing firms, marketplaces, insurers and software companies.
Should we block public AI tools entirely?
Usually not. Blocking without an alternative drives usage to personal devices where you have no visibility. A better approach is to provide an approved tool, block the riskiest categories of data, and monitor for unsanctioned services.
Closing both gaps
Delana Technologies helps organizations bring shadow AI under governance and strengthen identity verification across employees, vendors and customers, so the data that fuels fraud stays inside and the identities that commit it stay out. To start with an assessment of your AI use and identity controls, call 239.414.5126 or contact us.
Sources: IBM, Cost of a Data Breach Report 2025 and accompanying press release (July 30, 2025); public reporting on Samsung’s generative AI restrictions (May 2023).