Picture this. An account manager is putting together a loan summary for a client review. The draft needs polish, so she opens a popular AI chatbot, pastes in the client's name, income, credit history, and account balances, and asks for help cleaning up the language. Done in three minutes. Summary looks great.

Also done: she may have just shared that client's protected financial data with a third-party AI vendor her firm has never vetted, never contracted with, and has zero visibility into.

Shadow AI in financial services is no longer a hypothetical risk category. The Verizon 2026 Data Breach Investigations Report, published May 20, found that employee use of unapproved AI tools, what the industry calls shadow AI, tripled over the past year. It now affects 45% of the workforce across industries studied. For financial services firms navigating the Gramm-Leach-Bliley Act (GLBA), that number is not just an IT problem. It is a compliance exposure that most firms have not yet addressed.

What Shadow AI Actually Is

Shadow AI is any AI tool an employee uses without the knowledge or approval of the IT or compliance team. ChatGPT. Microsoft Copilot accessed through a personal account. Claude. Gemini. A browser-based summarization plugin. These tools are free, fast, and genuinely useful. That is exactly why employees adopt them without asking.

Most firms have acceptable use policies that technically cover this. Most employees either do not know about those policies or do not think they apply to a quick AI writing assist on their lunch break.

The Verizon DBIR flagged shadow AI as a primary driver of a new and difficult-to-detect data leakage vector. Unlike a phishing attack or a ransomware deployment, shadow AI exposure leaves no obvious incident to investigate. The data just leaves. Quietly. Often repeatedly.

Why Shadow AI Financial Services Risk Is Different from Other Industries

Financial firms operate under GLBA, which requires them to maintain a written information security program, protect customer financial information against unauthorized disclosure, and ensure that any vendors handling customer data are contractually bound to protect it.

An AI chatbot accessed through an employee's free personal account is not a vetted vendor. It is not covered by your information security program. The data pasted into it is, depending on the provider's privacy terms and how the model is trained, potentially retained, used for model improvement, or accessible to the provider's support staff.

BakerHostetler's 2026 Data Security Incident Response Report (financialcontent.com) noted that vendors were responsible for 25% of the incidents it analyzed. Shadow AI extends that vendor risk to tools that do not even appear on your vendor list. They are invisible to your procurement team, your security tools, and your annual GLBA audit.

That gap is real.

The Numbers From the Verizon Report Are Worth Taking Seriously

The 2026 DBIR drew on data from thousands of confirmed incidents. A few findings that stand out for financial services:

The human element was involved in 62% of all breaches studied. Not malware. Not sophisticated exploits. People making decisions, sometimes well-intentioned ones, that created openings.

Third-party breaches jumped 60% year over year and now account for 48% of all incidents in the report. That figure includes both traditional vendor compromises and the newer category of unmanaged employee-adopted tools.

Vulnerability exploitation overtook credential theft as the top breach entry point for the first time in the report's 19-year history. Median time to patch is now 43 days. Software flaws are going unaddressed longer, and attackers are moving faster with AI-assisted tools that can turn a known vulnerability into a working exploit in hours.

The picture that emerges is one where both the perimeter and the human layer are under more pressure simultaneously.

A Contrarian Take: The Risk Is Not the AI, It Is the Gap in Policy

Here is something most cybersecurity commentary gets wrong: shadow AI is not an AI problem. It is a governance gap problem.

Employees are going to use AI tools because those tools help them do their jobs better. Fighting that is like fighting the adoption of cloud storage. The organizations that will create genuine security exposure are the ones that respond with "no AI tools allowed" policies that nobody follows, creating shadow usage plus plausible deniability all the way up the org chart.

The financial firms handling this well are doing the opposite: auditing which AI tools employees are already using, evaluating which ones can be approved for use with properly structured data (meaning no raw customer records pasted in), deploying a sanctioned AI platform with enterprise data protections, and updating their GLBA information security documentation to reflect how AI tools are now governed.

That is a realistic and achievable response. It does not require eliminating productivity gains. It requires knowing where the data goes.

Why This Matters for Financial Services Firms

GLBA enforcement is not passive. Regulators have made clear in 2026 guidance that they treat cybersecurity as a compliance failure, not just an IT issue. If a shadow AI incident surfaces customer data and your firm cannot demonstrate that you had a reasonable program in place to prevent it, the exposure goes beyond the breach itself.

Financial services also face heightened reputational stakes. A local credit union or investment advisory firm that loses client financial data to an unsanctioned AI tool faces a trust problem that a press release and a credit monitoring offer cannot fully repair. Your clients are with you because they believe you protect their money and their information. Shadow AI puts both at risk in a way that is entirely preventable.

What to Do About It

A few concrete starting points:

Start by finding out what your employees are actually using. Most firms are surprised by the answer. Simple IT monitoring tools can surface this quickly.

Evaluate and approve a small set of AI tools that meet your security requirements. Sanctioned access is better than suppressed but continued shadow usage.

Update your acceptable use and GLBA information security policies to explicitly address AI tools. Include what categories of data can never be input into any AI tool, customer names, account numbers, transaction history, income data, and make the prohibition clear and simple.

Train staff on why this matters. The account manager pasting client data into a chatbot is not malicious. She is trying to do her job well. She needs to understand the specific risk, not a lecture about security policy.

At Mytec, we help financial services firms close exactly these kinds of governance gaps: building GLBA-aligned security programs that account for the real way employees work today, including the AI tools they have already adopted.