Mark McLaughlin, Product Marketing Manager | Aurascape
October 27th, 2026

Introduction

Panasonic had expressly prohibited an external AI service. A project member still uploaded personal information about approximately 334,000 people to it, according to the company’s October 23 disclosure. 

The incident exposes a fundamental gap in enterprise AI security: prohibiting an application does not necessarily prevent employees from using it. Written policies must translate into technical enforcement that stops unauthorized AI usage and sensitive-data transfers before they occur. Employees need clear guidance, and security teams need controls that can act on it. 

That is the job of AI Usage Control: discovering the AI people use and enforcing policy on both access to those services and the interactions within them. The decision depends on the application, account, activity, and data involved. That requires decoding the AI interaction itself so policy can distinguish an approved task from a prohibited transfer within the same application. 

What Panasonic disclosed

The incident occurred during an HR-system implementation involving Panasonic Operational Excellence and its subsidiary Panasonic Digital. Employee and former-employee records included employee numbers, addresses, phone numbers, and email addresses, but no names. Family emergency-contact records included names and contact details. 

Panasonic attributed the upload to a project member’s policy violation rather than an external attack. It reported the incident to Japan’s privacy regulator on September 29. After discovery, the project member deleted the AI chat history and relevant data from the work computer. The exposure was the upload to an unauthorized provider. At disclosure, Panasonic had confirmed no public exposure or concrete harm from misuse. 

The AI service, account type, connection path, and existing technical controls remain undisclosed. The notice confirms that a prohibited upload occurred, but does not establish whether technical access restrictions were in place, which connection path was used, or why the transfer was possible. 

Start with access to the AI tool

If a service is expressly prohibited, access policy should stop employees from using it on the paths the organization can control. That is the first opportunity to prevent exposure, before inspecting a prompt or deciding whether an attachment contains sensitive data. 

Keeping that boundary current requires finding AI tools beyond the familiar names. The long tail includes emerging applications, specialized assistants, and AI features embedded in trusted software and websites. A list maintained by hand leaves security teams researching new tools while employees are already trying them. 

Discovery needs to lead to risk assessment. How does the provider retain submitted data, use it for training, or share it with third parties? Current risk scores and the attributes behind them should feed access policies that automatically restrict identified tools based on risk and approval status. 

Account and entitlement are part of the decision

For permitted tools, approval needs to include the account and license in use. A company may approve an application through its managed enterprise tenant. A personal account in the same application can fall outside that approval. Entitlement identifies the subscription or license, such as a free, individual paid, team, or enterprise plan.

The website may look the same, while protections differ by service, plan, settings, and contract. Security teams should check whether submitted data may be used for training, how long it is retained, where it is stored and processed, and which administrative controls the company can enforce.

Those distinctions should inform policy within the AI interaction. Identity, group, and department also provide context: an HR analyst and a marketing employee may use the same service while handling different information. Permission to access an HR dataset does not automatically include permission to upload it, even through an approved enterprise account.

These are requirements to evaluate in your own environment. Panasonic has not disclosed which account or entitlement was involved in its incident.

Understand what happens inside the application

Consider a hypothetical HR analyst preparing a system migration. The analyst uses the company’s enterprise AI account to help write data-validation formulas. The task could be completed with column descriptions and synthetic examples. Uploading identifiable employee records would create a different risk, even within the same approved application and account. 

Traditional secure web gateways, cloud access security brokers, and data loss prevention tools provide important destination, application, and content controls. Their value depends on what they can inspect and enforce along the path in use. Recognizing an AI application does not establish that a control can decode its conversations or distinguish the activities within it. 

AI interactions can use streaming connections and encoded message formats that require application-specific decoding. Effective AI Usage Control needs to decode and interpret the interaction itself, including the account, activity, and sensitive content, then apply policy before a prohibited transfer completes. Security teams should test that capability directly, rather than infer it from an application appearing in an inventory. 

Enforce protection policy before submission

Access policy establishes which tools and accounts people may use. Protection policy governs the information and actions within permitted interactions. An alert after an upload can help the response, but it cannot undo the submission.

For the HR analyst, that could mean allowing help with a formula while stopping an employee-data upload. Depending on the sensitivity of the information and the supported interaction, policy can allow the request, redact sensitive text, or block the transfer entirely. For sensitive file uploads, blocking the transfer may be the appropriate action rather than attempting to modify the file. Warnings and coaching can direct the employee toward an approved account or a safer way to complete the work.

Disabling training does not authorize the upload

Training settings do not replace that decision. Panasonic relayed the provider’s response that training was disabled, while deleted data would be retained for 30 days under restricted access. The notice said retention would end on October 25; it was not independent confirmation of completed deletion. 

These protections are important, but they govern what happens after the provider receives information. Disabling training does not prevent an unauthorized upload. The organization still needs to decide whether the information may leave and enforce that decision before submission. 

Test the workflows your people use

Use synthetic sensitive data and representative accounts to test both prohibited access and permitted work. Ask the team to demonstrate four things: 

• Discovery and access: Identify long-tail AI use, explain the risk assessment, and show that configured policy blocks prohibited or high-risk tools on the tested paths. 
• Account-aware policy: Distinguish approved enterprise accounts and entitlements from unapproved use, with interaction policies that reflect user and group context. 
• Inline prevention: Inspect the relevant prompts, pasted text, and attachments. Stop restricted data before the destination receives it while allowing a permitted task to proceed. 
• Evidence: Establish which account, activity, policy, and enforcement action were involved, with appropriate restrictions on access to sensitive records. 

Record any paths outside the control’s coverage. A successful demonstration should show both that the restriction holds and that people can still finish approved work. 

Putting AI Usage Control to work with Aurascape

Aurascape brings these requirements together to support three practical outcomes. 

Prevent unauthorized AI use. Automated discovery identifies long-tail AI applications, while risk scoring evaluates their data practices and security risks. Configured access policies can automatically restrict identified tools according to risk and approval status on covered paths, before employees submit information. 

Protect data within approved AI. Aurascape decodes supported AI interactions inline, bringing account entitlement, user context, application activity, and sensitive content into the policy decision. Depending on the interaction and configured policy, it can allow, coach, notify, redact, redirect, or block, helping employees complete useful tasks while preventing unauthorized transfers. 

Investigate and demonstrate control. Aurascape preserves evidence of AI activity and enforcement decisions. Auri™, Aurascape’s natural-language governance assistant, helps authorized security, HR, legal, and compliance teams surface risky activity and investigate potential incidents through plain-language questions, with role-based access to the information relevant to their responsibilities. 

The lesson from Panasonic is straightforward: an AI policy is only as effective as the controls that enforce it. Organizations need to know which AI tools are being used, understand what happens inside them, and stop prohibited actions before sensitive information leaves. 

See how Aurascape enforces AI policies and prevents unauthorized sensitive-data transfers in AI interactions before they complete.