Comparing SPII vs PHI and PII: A Sensitive Information Guide

Key Takeaways:

• PII, PHI, and SPII are not interchangeable. PII identifies individuals. PHI covers identifiable health data under HIPAA. SPII is a high-risk subset of PII (Social Security numbers, passports, biometrics) that carries the heaviest penalties if exposed.
• These categories overlap more than most teams realize. Vaccination records mix PHI and PII. Payroll forms contain SPII and PII. Without context-aware classification, those overlaps go unnoticed and spread across SaaS tools and AI integrations.
• AI has rewritten the risk equation. When employees paste reports into Copilot or ask ChatGPT to summarize client files, sensitive data can end up in model memory, logs, and later prompts. Traditional pattern-matching DLP tools can’t keep up.
• Classification is no longer a checkbox exercise. With sensitive data moving across clouds, SaaS platforms, and AI workflows, organizations need active classification that understands what data means, where it lives, and how it’s being used.
• You can’t protect what you can’t see. Visibility and context are essential, especially when AI tools process and remember data faster than traditional security can track.

Sensitive data has never been more vulnerable, or more misunderstood. As data moves across clouds, SaaS platforms, and now into AI assistants like Copilot, ChatGPT, and Claude, organizations are realizing that not all “personal data” is created equal.

You’ve probably heard terms like PII, PHI, and SPII, and maybe you’ve used them interchangeably. But they relate to very different categories of information, and confusing them could lead to compliance problems, misclassified data, and sensitive data being leaked to AI.

This guide breaks down each category, shows where they overlap, and explains why understanding those differences matters for protecting your company’s most sensitive assets.

Why These Definitions Matter More Than Ever

A decade ago, these data categories mostly resided in compliance manuals: PII under the GDPR and CCPA, PHI under HIPAA, and so on. But with the explosion of AI tools, sensitive data isn’t sitting quietly and motionless in databases anymore. It’s being pasted into prompts, summarized by assistants, handled by agentic AI, and stored in logs that security teams rarely see.

That shift means data classification is no longer a checkbox exercise, but an active, ongoing defense strategy. CISOs and compliance teams need to know exactly what type of sensitive data is being used, where it’s stored, and who (or what AI tool) has access to it.

The stakes are no longer hypothetical. Texas secured a settlement of more than $1 billion against a major technology company under its state privacy law in 2026, and California’s privacy agency broke its own settlement record with a $2.75 million action against a streaming company over opt-out failures. Misclassifying a dataset, or missing that it qualifies as SPII instead of ordinary PII, is turning into a very expensive mistake.

Modern data security governance platforms can automate this process. They use context-aware intelligence to classify SPII, PHI, and PII across all repositories and AI interactions, enabling you to detect risks well before data leaves your control.

Defining PII, PHI, and SPII

Once upon a time, sensitive data fit neatly into categories. PII meant personal info, PHI meant health data, and SPII meant the really private stuff. However, with GenAI assistants now handling emails, reports, and code, those lines have begun to blur. 

Let’s break down what each type actually means and why keeping them straight has never been more important.

What is PII (Personally Identifiable Information)?

PII refers to any data that can identify an individual, directly or indirectly. Think names, email addresses, phone numbers, IP addresses, or government IDs. PII is governed by privacy laws like GDPR, CCPA, and LGPD.

While most organizations have policies around PII, AI tools are remixing the risk landscape. When employees upload reports or prompt Copilot to “summarize all client files,” those files often contain PII that may be exposed to AI model memory or logs.

The takeaway is this: PII might seem harmless on its own, but when combined with other data points or shared through AI workflows, it can easily cross into SPII or PHI territory.

What is PHI (Protected Health Information)?

PHI is any health-related data tied to an identifiable person. Under HIPAA, this includes medical histories, laboratory results, insurance claims, and appointment records.

Today’s risk is that healthcare teams and HR departments are increasingly relying on AI to generate summaries, track wellness programs, or process claims. If that AI assistant has access to PHI—especially outside HIPAA-compliant systems—the organization faces regulatory and reputational exposure.

Example: An AI tool summarizing a spreadsheet of patient test results might store that information in its history. Without oversight, that data can resurface in unrelated queries.

The compliance bar is about to get higher, if not there already. HHS proposed the most significant overhaul of the HIPAA Security Rule since 2003, which would make multi-factor authentication, encryption, network segmentation, and 72-hour breach notification mandatory for every covered entity regardless of size. 

The rule is still pending finalization, but OCR has already begun civil enforcement of 42 CFR Part 2 as of February 2026, extending HIPAA-level penalties to substance use disorder treatment records for the first time. For any organization running PHI through AI tools, the direction of travel is unmistakable even before the Security Rule update lands.

What is SPII (Sensitive Personally Identifiable Information)?

SPII takes PII a step further. It includes information that could cause serious harm if disclosed, such as Social Security numbers, passport details, financial account numbers, or biometric data.

This category is recognized by frameworks like NIST SP 800-122 and U.S. federal privacy laws. Breaches involving SPII often trigger strict notification requirements and carry the heaviest penalties.

When it comes to AI, SPII can be exposed in subtle ways. In one example, an employee may paste a payroll record into Claude or share tax information through a AI-driven workflow.

In short, SPII = PII, which can do real damage if mishandled. That definition keeps expanding too.

State legislatures are actively adding categories like neural data, inferred data, and financial credentials to their sensitive personal data definitions, which means SPII in 2026 covers considerably more ground than it did even two years ago.

SPII vs PHI vs PII: A Quick Comparison

Where PII, PHI and SPII Overlap

PII, PHI, and SPII aren’t isolated boxes, as they do overlap. Every piece of PHI, for example, is also PII. And much of SPII, like a driver’s license number, can qualify as PII (or even PHI) if used in a medical record.

Here’s where things get messy in practice:

• HR teams may store vaccination records (PHI) in employee files (PII).
• Finance departments process payroll forms that contain both Social Security numbers (SPII) and addresses (PII).
• Marketing teams use AI to craft personalized outreach, often combining multiple data types without being aware of it.

Without precise (and context-aware) classification and continuous monitoring, those overlaps make it easy for sensitive information to spread unnoticed across SaaS tools and AI integrations.e overlaps make it easy for sensitive information to spread unnoticed across SaaS tools and GenAI integrations.

How GenAI Changes the Risk Equation

Traditional DLP tools look for patterns such as credit card numbers, SSNs, or regular expression (regex) matches. But AI rewrites that script with a huge red marker.

AI tools not only process data, but they also understand it, remember it, and can accidentally reuse it.

For example:

• A Copilot instance trained on shared drives might inadvertently learn sensitive customer data.
• A ChatGPT Enterprise account could summarize internal documents containing PHI, and parts of that data could reappear in later prompts.
• AI-generated drafts in Claude can embed fragments of SPII or PHI that evade legacy scanners. 
• Security researchers demonstrated that a compromised AI agent in a major cloud platform could exploit default permission settings to pivot from a single customer environment into far more sensitive internal systems.

This is why classification alone isn’t enough. You need context: what the data means, where it lives, and how it’s being used.

None of this is happening on sanctioned tools alone. Nearly half of employees using AI at work are doing so through personal accounts, completely outside their employer’s visibility, and prompt volume to these tools has grown sixfold in a single year. Add agentic AI and connector-based integrations into the mix, where a single compromised connection can reach far beyond its intended scope, and exposure for SPII, PHI, and PII stops looking like a data problem and starts looking like an identity and access problem.

How Concentric AI Protects SPII, PHI, and PII

Concentric AI’s Semantic Intelligence™ platform discovers, classifies, and protects sensitive data wherever it lives across the Concentric AI’s Semantic Intelligence and Semantic DLP™ platforms discover, classify, and protect sensitive data wherever it lives across the enterprise—structured or unstructured, cloud or on-prem.

Semantic Intelligence’s key capabilities include:

• Classification with context: Automatically identifies PII, PHI, and SPII—even in unstructured documents, chat logs, and AI-generated files.
• AI-Aware risk detection: Flags sensitive data embedded in prompts, training datasets, or AI summaries.
• Access governance: Detects oversharing, misconfigured permissions, and AI integrations that increase exposure.
• Automated remediation: Supports deletion, archiving, masking, or blocking data to meet regulatory and organizational standards.

Concentric AI’s approach goes well beyond discovery, providing the continuous governance needed to keep data safe in an AI-driven workplace. Organizations get visibility into how AI is being used in their organization, identifies sensitive data before it ever reaches prompts, and helps apply and enforce the guardrails needed to keep data secure.