Financial services AI security is now a GLBA and SEC issue

At a glance

What this is: Cyberhaven argues that rapid AI adoption in financial services is creating a measurable governance gap because regulated data is moving through shadow AI, endpoint agents, and embedded AI features faster than firms can control it.

Why it matters: This matters to IAM and security teams because AI use now intersects with access control, data handling, auditability, and disclosure obligations, especially where employees and agents handle nonpublic financial data outside approved workflows.

By the numbers:

• Cyberhaven says there is a 17x difference in AI adoption between the most aggressive financial services firms and the most cautious ones.

Context

Financial services AI security is fundamentally a governance problem: firms can adopt AI faster than they can prove where regulated data moved, who touched it, and what controls were active. In regulated environments, the security question is not whether AI can be used, but whether its use remains inside the firm's control and disclosure boundary.

The regulatory pressure comes from two directions. The GLBA Safeguards Rule requires monitoring, logging, access control, and breach notification for customer information, while SEC rules require material incident disclosure and annual reporting on cybersecurity risk management. That makes AI use an issue for IAM, data security, and governance teams, not just AI builders.

Key questions

Q: How should organisations govern shadow AI without blocking legitimate use?

A: Start with approved-use policy, tool inventory, and data classification. Then require that any AI system handling internal information has named owners, logged access, and defined credential paths. The goal is not prohibition, but visibility and control. If a tool cannot be inventoried or monitored, it should not process sensitive data.

Q: Why do endpoint AI agents create regulatory risk for financial firms?

A: Because they inherit a user's access and can move or transform data locally without the same visibility as a standard application session. That makes them a high-risk delegated access path for nonpublic personal information, especially when firms need to prove where data went and who touched it.

Q: What are the signs that AI use is outside a firm's control boundary?

A: Common signals include personal-account usage, AI features hidden inside approved software, data submissions that never appear in central logs, and agent activity that cannot be traced back to a governed workflow. If the firm cannot reconstruct the session and the data path, control is already failing.

Q: When do AI data exposure events become a GLBA or SEC accountability issue?

A: When the firm cannot show what customer data was processed, whether monitoring and logging were active, and whether the exposure reached the threshold for breach notification or material disclosure. At that point the issue is not only technical loss, but reporting and board accountability.

Technical breakdown

Shadow AI and personal accounts create ungoverned data paths

Shadow AI arises when employees use AI tools without security review, often through personal accounts that sit outside corporate visibility. The problem is not just unsanctioned software, but unsanctioned identity and data paths, because the user session is no longer tied to managed controls, logging, or retention rules. In financial services, that creates a direct path for nonpublic personal information to leave governed workflows. Endpoint and network controls often miss this because the session looks like normal cloud usage. The governance failure is visibility at the point of use, not policy language.

Practical implication: Security teams need enforceable visibility into AI sessions that originate outside managed identities and approved application paths.

Endpoint AI agents extend user privilege into local data movement

Endpoint AI agents operate on the employee device, where they can read files, move content, and act with the privileges of the launching user. That means the agent inherits human access, but its actions may be faster, broader, and less observable than a person working manually. From an identity perspective, the agent becomes a transient extension of user authority without a clearly bounded lifecycle. From a data perspective, this makes endpoint AI a high-risk transfer mechanism for customer records and internal material. Traditional perimeter monitoring is weak here because the activity is local and often encrypted in transit.

Practical implication: Treat endpoint AI agents as delegated access paths and restrict what data they can reach from the user session they inherit.

Embedded AI features blur the boundary between approved software and exfiltration

Many AI capabilities now sit inside tools employees already trust, which means risky data movement does not look like a separate exfiltration event. The user may believe they are staying inside an approved application while content is silently sent to an AI feature with its own processing and retention model. That makes classification, logging, and disclosure harder, because the risk is hidden inside ordinary business software. For regulated firms, the core control issue is not whether the application is sanctioned, but whether its AI functions are governed as a distinct data path. Without that distinction, policy and reality diverge.

Practical implication: Inventory AI features inside approved software and govern them as separate data flows, not as ordinary application usage.

Threat narrative

Attacker objective: The practical objective is not always external theft; in many cases the result is uncontrolled disclosure, loss of auditability, and regulatory exposure around regulated financial data.

1. Entry occurs when employees submit customer or firm data into unsanctioned AI tools, personal accounts, or embedded AI features without review.
2. Credentialed access expands as endpoint AI agents and approved applications inherit the user's privileges and can act on sensitive files and content.
3. Impact follows when the firm cannot reconstruct where regulated data went, who touched it, or whether the event is material for GLBA and SEC reporting.

Breaches seen in the wild

• DeepSeek database exposure 2025: An unauthenticated DeepSeek ClickHouse database exposed over a million log lines with plaintext chat history and API keys in 2025.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI security in financial services is now a disclosure problem as much as a data control problem. The article shows that AI use can trigger GLBA monitoring expectations and SEC materiality assessment at the same time. That means firms are no longer just governing model use or employee productivity tools. They are governing evidence, because if they cannot reconstruct the data path, they cannot reliably defend compliance.

Shadow AI is a lifecycle and identity problem, not just an application inventory issue. Personal accounts, endpoint agents, and embedded AI features all create sessions that sit partly outside enterprise control. In identity terms, this is delegated access without a clean governance boundary, which is exactly where accountability weakens. Practitioners should read this as a signal that inventory alone is insufficient unless it is tied to enforceable identity and data controls.

Data lineage is becoming the operational proof layer for regulated AI. Financial services firms need records that show where nonpublic data moved, which identity path carried it, and which controls were active at the time. That is a stronger requirement than simple logging, because regulators care about reconstructability, not just event capture. The practical conclusion is that auditability must be designed into AI workflows from the start.

Endpoint AI agents create a new form of privilege amplification. When the agent inherits human access and acts locally, it can move faster than review processes can follow. This is where access governance assumptions start to break, because the system now makes decisions and transfers data within a session that looks human but behaves mechanically. Teams should treat agent activity as governed delegated privilege, not ordinary user behaviour.

From our research library:

• 28% of secrets incidents now originate outside code repositories, in Slack, Jira, and Confluence, and are 13% more likely to be categorised as critical than code-based leaks, according to the State of Secrets Sprawl 2026.
• Only 44% of developers are reported to follow security best practices for secrets management, exposing a significant developer behaviour gap, according to the State of Secrets in AppSec.
• Read next: Enterprise AI Copilot Security Guide

What this signals

AI governance debt: financial services firms are accumulating AI usage faster than they are building the evidence layer needed for GLBA and SEC accountability. The gap is not only in policy but in reconstructability, because regulators will expect firms to explain where regulated data went and which controls were active.

The practical signal for IAM and security leaders is that identity visibility now has to extend into AI sessions, endpoint agents, and embedded features. If those paths are not tied to governed access, incident response and disclosure workflows will fail under pressure.

For practitioners

• Inventory all AI entry points Catalog standalone GenAI tools, embedded AI features, personal-account usage, and endpoint agents so regulated data paths are visible before you set policy.
• Enforce data controls at the point of submission Block or alert when nonpublic personal information, account data, or sensitive internal content is entered into AI tools, rather than relying on network-only inspection.
• Separate approved software from approved AI processing Treat AI functions inside sanctioned applications as distinct data paths with their own logging, retention, and approval requirements.
• Tie agent activity to governed identity paths Restrict endpoint AI agents to the minimum data scope of the launching user and require traceability for every action they take with regulated information.
• Build audit-ready evidence for GLBA and SEC Document what data each AI tool processes, what controls apply, and how incidents are assessed for notification and materiality decisions.

Key takeaways

• Financial services AI risk is no longer limited to model safety or productivity governance, because regulated data now moves through tools that may sit outside approved control paths.
• The core failure is evidentiary: firms may not be able to prove where customer data went, who accessed it, or whether the right controls were active when AI was used.
• The immediate response is to treat AI entry points as governed data flows, then tie those flows to identity, logging, and disclosure-ready records.

Key terms

• Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.
• Data Lineage: The record of how data moves across systems, applications, and workflows. In security operations, lineage shows where sensitive data propagates, which identities touch it, and how a compromise could spread across connected environments.
• Desktop AI Agent: A desktop AI agent is an AI-enabled tool that runs in a local workstation environment and can act on behalf of a user across applications, files, and services. Unlike browser-only AI use, its behaviour may span endpoints, network paths, and integrated productivity tools, making visibility and policy enforcement harder to centralise.
• Materiality assessment: A process for deciding which issues matter enough to influence strategy, reporting, and control priorities. In this context, it is useful when it is refreshed by current risk signals rather than treated as a fixed annual exercise.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity, and machine identity security. It helps practitioners connect access control, lifecycle governance, and evidence requirements across identity programmes.