PII Security in the Age of AI: Best Practices

Protect PII against AI-enhanced threats including credential stuffing and deepfakes. Learn essential security practices to avoid costly breaches and regulatory penalties in AI environments.

Understanding Personally Identifiable Information (PII)

Personally identifiable information (PII) is any data that identifies a specific individual or can be combined with other information to identify someone. PII stands for personally identifiable information: the data elements that distinguish you from everyone else. Names, Social Security numbers, email addresses, phone numbers, and biometric data like fingerprints all qualify as PII.

No single standard defines PII consistently across all jurisdictions and industries. The PII you must protect depends on which jurisdictions regulate your operations and what type of data your systems process. Apply the most restrictive PII definition governing your operations. 

NIST defines PII as information distinguishing or tracing identity such as name, SSN, biometrics, alone or combined with other personal information. GDPR Article 4(1) expands this to "any information relating to an identified or identifiable natural person," explicitly including online identifiers like IP addresses, cookies, and device fingerprints. CCPA §1798.140 takes the most granular approach, defining biometric information as "keystroke patterns or rhythms, gait patterns or rhythms, and sleep, health, or exercise data that contain identifying information."

AI fundamentally expanded what constitutes PII. The keystroke dynamics your fraud system analyzes? Biometric data under CCPA. The behavioral risk scores your access control platform generates? Personal information under CCPA §1798.140(o)(1)(K), which uniquely categorizes AI-generated inferences as personal information, including profiles reflecting consumer preferences, psychological trends, and behavior. New PII categories including faceprints and biometric templates that your systems must safeguard differently than traditional identifiers.

Why PII Security Matters in Cybersecurity

U.S. organizations face average breach costs of $9.36 million: nearly double the $4.88 million global average. Beyond direct costs, PII compromise creates a multiplier effect that compounds financial and operational damage.

U.S. organizations pay 194% of the global average for data breaches: $9.48 million versus $4.88 million worldwide. Those numbers jumped 10% year-over-year in 2024, marking the steepest increase since the pandemic period.

When you delay response, costs multiply exponentially. Breaches exceeding 200 days to identify and contain can cost $5.46 million. The longer attackers persist in your environment, the more you pay. Regulatory scrutiny intensifies. Customer turnover accelerates. Operational disruption compounds.

The 2024 Verizon Data Breach Investigations Report analyzed 30,458 security incidents and 10,626 confirmed breaches, a record dataset representing a two-fold increase in reported incidents. Breaches compromising PII weren't an edge case. They followed the predominant pattern across every major industry sector.

Those breach statistics measure yesterday's threat landscape. AI fundamentally changed what you're defending against and what constitutes PII in the first place.

Key Risks and Threats to PII

Three primary attack vectors target PII across enterprise environments: 

1. Unauthorized access accounts for the majority of external breaches. Attackers exploit weak authentication, unpatched vulnerabilities, and misconfigured systems to reach PII databases. Once inside your network, lateral movement techniques allow attackers to escalate privileges and access sensitive data stores. Human elements drive most breaches: stolen credentials, phishing, or misuse of access privileges.
2. Insider threats operate from trusted positions within your organization. Employees, contractors, and business partners with legitimate access can exfiltrate PII intentionally or accidentally expose data through negligence. Malicious insiders cause particularly costly breaches because they understand your security controls and know where valuable data resides.
3. Third-party vendors create extended attack surfaces beyond your direct control. When you share PII with cloud providers, payment processors, or analytics platforms, you depend on their security controls. Supply chain compromises targeting vendor systems provide attackers indirect access to your PII through trusted business relationships.

Understanding these foundational threats establishes the context for implementing core protection principles that address each attack vector systematically.

Core Principles of PII Protection

Five foundational principles govern effective PII security regardless of specific compliance framework or industry vertical.

1. Data minimization: Collect only the PII necessary for defined business purposes. When you don't store data, attackers can't steal it. Review collection practices quarterly and delete unnecessary PII. Minimization reduces both privacy and security risks by limiting exposure.
2. Purpose limitation: Use PII only for the specific purposes disclosed during collection. Processing PII beyond original intent requires explicit consent or legitimate grounds. Document every processing purpose and restrict system access accordingly.
3. Storage limitation: Retain PII only as long as necessary for legitimate business or legal requirements. Implement automated deletion policies based on retention schedules. Keep personal data in identifiable form no longer than necessary for its intended purpose.
4. Integrity and confidentiality: Protect PII through appropriate technical and organizational measures. Encryption, access controls, and security monitoring prevent unauthorized access and modification. These controls must address both accidental loss and deliberate attacks.
5. Accountability: Demonstrate compliance through documentation, audit trails, and governance processes. You must prove your controls work, not just claim they exist.

These principles form the baseline that AI-specific controls build upon, addressing both traditional and emerging PII security challenges.

How AI Has Impacted PII Cybersecurity Measures

AI didn't just accelerate the attacks you face. It created entirely new attack methods targeting your systems and expanded what constitutes personally identifiable information in ways traditional defenses can't address.

AI-Enhanced Attack Techniques Targeting PII

A 19% credential stuffing rate in 2025 means attackers run optimized authentication attempts against your login systems continuously. Machine learning algorithms test credential combinations harvested from previous breaches, optimizing patterns based on your system's responses. Even small organizations experience 12% attack rates, and successful credential stuffing bypasses your perimeter security entirely, delivering authenticated access to PII databases.

British engineering firm Arup lost $25 million in 2024 after deepfake video impersonated their CFO during a call. Your security awareness training taught employees to verify unusual requests via phone call. What happens when the phone call itself is synthetic?

The FBI documented $16.6 billion in losses from 859,532 complaints in 2024, representing a 33% increase from 2023. Federal agencies warned specifically that criminals use AI to craft highly convincing voice or video messages and emails. Verizon's research confirms 60% of phishing incidents are identity-based attacks, with 50% of users opening phishing emails within the first hour.

Threats to PII in AI Systems

Data poisoning threatens your AI systems directly. When you train models on compromised datasets, attackers can manipulate the AI making PII classification and access control decisions. According to NSA, CISA, and FBI joint guidance from May 2025, attackers identify maliciously modified training data as a primary method against AI systems. Your AI might be learning from attacker-controlled examples.

These attack methods compound the fundamental expansion of what constitutes PII in the first place.

AI Expansion of PII Categories

DHS's January 2025 update documents active deployment of facial recognition and biometric capture technologies with enhanced governance frameworks. Organizations must classify what is PII data versus non-personal information as AI systems create new categories: faceprints, biometric templates, behavioral patterns that require different security controls than name and SSN combinations.

CCPA explicitly regulates keystroke dynamics, gait analysis, voice prints, and health data from wearables. If your fraud system analyzes how users type, your endpoint security monitors mouse movements, or your physical security tracks walking patterns, you're processing biometric PII under California law.

California's regulation of AI-generated inferences expands the definition of personal information. When your recommendation engine creates behavioral profiles, your risk scoring platform predicts user actions, or your analytics system infers psychological characteristics, those algorithmic outputs constitute personal information under California Civil Code §1798.140(o)(1)(K). You're responsible for predictions your models generate, not just data you originally collected.

You can't protect against AI-enhanced attacks without understanding your specific compliance obligations, which vary dramatically based on jurisdiction and industry.

PII Compliance Considerations

Compliance isn't about checking boxes. It's about demonstrating accountability when regulators investigate your breach, and the financial stakes are substantial. Regulations define PII differently across jurisdictions, creating complex obligations for organizations operating globally.

GDPR Requirements

Article 32 requires pseudonymization, encryption, ongoing confidentiality, integrity, availability, and resilience of processing systems, as well as the ability to restore availability and access to personal data in a timely manner after incidents. You must respond to data subject access requests within one month, providing purposes of processing, categories of personal data, recipients, and retention periods.

EDPB Opinion 28/2024 established that deploying AI models creates controller obligations. Before you deploy any AI system processing personal data, you must ascertain whether vendors developed the model through lawful processing. You cannot claim ignorance about your AI vendor's training data sources. When you deploy AI based on legitimate interest, you must conduct a three-step legitimate interest assessment documenting necessity, balancing tests, and safeguards.

When you process special categories of personal data through AI systems, you need Article 9 exemptions under GDPR. According to European Parliament analysis and EDPB guidance, you cannot process sensitive personal data for bias analysis, even with good intentions, without meeting one of Article 9's limited grounds.

CCPA ADMT Requirements

You have less than 12 months to implement automated decision-making technology (ADMT) transparency requirements. California businesses must provide meaningful information about ADMT logic, descriptions of likely outcomes for individual consumers, and functional opt-out mechanisms. These requirements take effect January 1, 2026. Cybersecurity Services audit requirements and mandatory risk assessments take effect simultaneously.

The $7,988 per-violation penalty for intentional violations or those affecting minors turns noncompliance into eight-figure exposure for large-scale operations.

HIPAA's Requirements

HHS guidance on online tracking technologies requires you to configure user-authenticated webpages with tracking technologies to only use and disclose PHI in compliance with HIPAA Privacy Rule. Additionally, covered entities must enter into business associate agreements (BAAs) with tracking technology vendors when they disclose PHI. All ePHI collected through websites or apps requires Security Rule protections.

You need business associate agreements before disclosing PHI to tracking technology vendors. Your analytics platform, appointment scheduling system, and marketing automation tools all require BAAs when they process PHI. Appointment scheduling systems using third-party tracking that automatically transmit PHI to vendors require BAAs. Business associate security failures trigger six-figure penalties.

Risk analyses under 45 CFR § 164.308(a)(1)(ii)(A) must encompass all e-PHI you create, receive, maintain, or transmit. According to HHS guidance, this analysis must be regularly reviewed and updated as part of ongoing security management processes, not a one-time assessment.

Regulatory frameworks tell you what to protect. Understanding common mistakes helps you avoid the pitfalls that lead to compliance failures and breaches.

Common PII Security Mistakes and Challenges

Avoid these critical mistakes that expose PII in AI systems and create regulatory liability.

• Don't deploy AI systems without prompt injection defenses. Prompt injection attacks appear in OWASP's Top 10 for LLM Applications because attackers can manipulate AI systems to extract PII through carefully crafted inputs. When your AI chatbot processes user queries, customer support tickets, or external content, malicious instructions embedded in that content can trick the model into revealing sensitive data. Services like Prompt Security, a SentinelOne company, provide specialized detection for prompt injection attacks contextualized to your application's use case. It continuously evolves to counter new attack methodologies, such as those targeting PII extraction from AI systems.
• Don't assume de-identification protects PII in your AI training data. Models trained on "anonymized" datasets can leak identifying information through generated outputs or membership inference attacks. According to EDPB Opinion 28/2024, controllers deploying AI models must ascertain that vendors did not develop models through unlawful processing of personal data.
• Don't treat your AI-generated inferences as non-personal data. California explicitly regulates predictions derived from personal information. According to California Civil Code §1798.140(o)(1)(K), AI-generated inferences, including profiles reflecting consumer preferences, psychological trends, and behavior, are uniquely categorized as personal information.
• Don't deploy AI models without validating your training data compliance. EDPB Opinion 28/2024 establishes that controllers must ascertain whether vendors developed AI models through lawful processing and demonstrate compliance with GDPR Article 5(1)(a) before deployment. "We bought the model from a vendor" does not satisfy accountability obligations.
• Don't ignore that 19% credential stuffing rate hitting your systems. Multi-factor authentication stops this attack method immediately. When attackers succeed with credential stuffing, they gain authenticated access to PII databases, triggering $9.48M average breach costs.
• Don't fail to update your business associate agreements for AI services. Your analytics platform added AI features that process PHI differently than rule-based analysis. Your original BAA doesn't cover the new processing activities. According to EDPB Opinion 28/2024, controllers must ascertain that vendors did not develop AI models through unlawful processing.
• Don't overlook behavioral biometrics in fraud analysis. Your fraud platform analyzes typing patterns, mouse movements, and device interaction rhythms. That's biometric information under CCPA requiring specific notices, collection limitations, and retention policies.
• Don't assume GDPR legitimate interest covers AI processing automatically. Controllers deploying AI models based on legitimate interest must conduct three-step assessments demonstrating necessity, balancing tests, and appropriate safeguards.
• Don't operate AI systems without complete logging. When regulators investigate your breach, you need evidence demonstrating what your AI systems accessed, processed, and output. "The model made that decision" doesn't satisfy accountability requirements without audit trails.

Knowing what to avoid is only half the battle. The next section shows you how to implement integrated technical controls that protect PII throughout the AI lifecycle.

Best Practices for PII Security in AI Environments

Integrate five frameworks simultaneously: NIST Privacy Framework 1.1, NIST Cybersecurity Framework 2.0, CISA AI Roadmap, SANS Critical AI Security Guidelines, and jurisdiction-specific compliance (GDPR/CCPA/HIPAA). Technical controls and continuous monitoring determine whether you survive the breach.

Integrate NIST Privacy Framework 1.1 with Cybersecurity Framework 2.0

NIST's Privacy Framework 1.1 gives you five core functions designed for PII protection:

• IDENTIFY-P: Inventory every AI system touching PII, document data flows from collection through disposal, and map business context including third-party processors. Per NIST Privacy Framework 1.1, your risk assessments must account for AI-specific threats including model poisoning, training data compromise, inference leakage, and adversarial attacks.
• GOVERN-P: Establish privacy governance that integrates with cybersecurity risk management. Define roles covering AI model oversight, including accountability assessments before AI deployment. Your policies must address AI training data sourcing, third-party model evaluation, algorithmic decision review, and supply chain security.
• CONTROL-P: Implement disassociated processing to minimize PII exposure in AI operations. Process de-identified data for training when possible. Enforce data lifecycle policies that cover model training, inference, and retention separately.
• COMMUNICATE-P: Create transparency mechanisms explaining AI processing to data subjects. CCPA's ADMT requirements, effective January 1, 2026, mandate meaningful information about decision-making logic, descriptions of likely outcomes, and opt-out mechanisms.
• PROTECT-P: Deploy identity management, authentication, access control, encryption, and platform security controls designed for AI infrastructure. This includes model integrity verification to find tampering, protection against model theft and reverse engineering, version control and integrity checks, and access controls for model deployment with separation of duties.

• GOVERN, NIST CSF 2.0's sixth function, provides organizational-level cybersecurity risk management. According to NIST's announcement, the Privacy Framework 1.1 was built for direct integration with Cybersecurity Framework 2.0, enabling organizations to implement both frameworks in parallel.

Integrate SANS Critical AI Security Guidelines

SANS provides six operational control categories specifically for AI systems. What is PII data that requires protection? These controls help you identify and secure it throughout the AI lifecycle:

1. Input Validation and Sanitization: Protect against prompt injection, data poisoning, adversarial inputs that could extract PII
2. Model Security: Secure models from tampering, theft, unauthorized access; protect model weights and architecture. This means protecting weights, architecture, and training pipelines from tampering and theft. According to NSA, CISA, and FBI joint guidance, attackers who steal your model can reverse-engineer training data, potentially extracting PII embedded in model parameters. Version control and integrity checks prevent unauthorized model modifications that could disable PII protections.
3. Output Controls: These validate and monitor what your AI systems return to users. Models occasionally leak training data through generated outputs. According to SANS Critical AI Security Guidelines v1.1, output controls validate and monitor AI-generated content to prevent PII leakage before responses reach users.
4. Access Controls: These controls require role-based access control with strong authentication. Your AI platforms need separate permissions for training data access, model deployment, inference queries, and administrative functions. According to NIST Privacy Framework 1.1, separation of duties prevents single-account compromise from exposing all AI operations
5. Data Protection: Safeguard training data and operational data throughout AI lifecycle: collection, preprocessing, training, inference, storage, requires controls beyond traditional database security. Training data stores need encryption at rest, in transit, and during processing. Inference data requires equal protection even when using "anonymized" inputs that might be re-identified.
6. Monitoring and Logging: Complete logging of AI inputs, outputs, and decisions enables security monitoring and compliance auditing. You can't investigate what you didn't log. Your SIEM must ingest AI platform telemetry alongside traditional security events.

Verify Every AI System Request with Zero Trust

Zero Trust architecture is a critical security principle for protecting PII in AI systems. CISA's AI Security Roadmap emphasizes that AI systems require the same Secure by Design principles as any software system, protected from cyber threats throughout their entire lifecycle with security assessment before deployment, continuous monitoring during operations, and supply chain security for AI components and training data.

NIST Cybersecurity Framework 2.0 incorporates Zero Trust principles through its PROTECT (PR) function, specifically requiring strong identity and authentication/access control mechanisms. When combined with the NIST Privacy Framework 1.1, organizations must implement identity management and authentication controls specifically designed for AI systems handling personally identifiable information.

NSA, CISA, and FBI joint guidance on AI data security reinforces that malicious actors use compromised credentials and supply chain vulnerabilities to access AI training pipelines and inference systems.

Verify every access request regardless of network location. Your AI training infrastructure shouldn't trust requests from corporate networks by default. Implement least-privilege access for service accounts running AI workloads. Segment AI systems processing different PII categories into separate trust zones. Deploy multi-factor authentication to counter credential stuffing attacks.

You must monitor continuously throughout the operational lifecycle, not just during deployment. AI system behavior drifts over time: model decay, data distribution changes, and adversarial adaptation change your risk profile. Response windows compress from weeks to minutes. According to NIST Cybersecurity Framework 2.0, continuous monitoring (DE.CM) and adverse event analysis (DE.AE) are essential functions.

Address Supply Chain Security for AI Components and Training Data

Third-party AI services expand your attack surface beyond infrastructure you control. According to NSA, CISA, and FBI joint guidance (May 2025), AI training data supply chains represent a primary attack method where attackers inject maliciously modified "poisoned" data into AI training sets.

Conduct security assessments before deploying AI systems, as required by CISA's AI Security Roadmap. Your vendor security questionnaires need questions about training data provenance, model development environments, and security controls protecting model artifacts. Document the entire AI supply chain including datasets, pre-trained models, APIs, and inference platforms.

Establish business associate agreements or data processing agreements covering AI vendors before they process PII. Your contracts must specify data handling requirements, security controls, incident notification timelines, and audit rights. Under GDPR Article 28, processor agreements must cover the subject matter, duration, nature and purpose of processing, types of personal data, categories of data subjects, and obligations and rights of the controller. HIPAA business associate agreements must address Security Rule and Privacy Rule requirements. CCPA service provider contracts require appropriate terms and verification of data broker registration status.

These technical controls and frameworks create the foundation for PII protection, but implementation requires autonomous capabilities that execute at the speed of AI-enhanced threats.

PII Security in Cloud and Hybrid Environments

Cloud and hybrid infrastructures create unique PII security challenges that traditional perimeter defenses can't address. Shared responsibility models, dynamic resource allocation, and multi-tenant architectures require specialized controls.

• Shared responsibility gaps: Cloud providers secure infrastructure, but you remain responsible for data protection. Misconfigured storage buckets, overly permissive access policies, and unencrypted data stores expose PII to unauthorized access. Cloud-based breaches typically stem from misconfigurations rather than provider vulnerabilities.
• Data residency and sovereignty: PII stored across multiple cloud regions must comply with jurisdiction-specific regulations. Some frameworks restrict transfers outside specific geographic boundaries without adequate safeguards. Regulations apply to residents regardless of where you process their data. Map data flows across cloud regions and implement localization controls where required.
• Container and serverless security: Ephemeral compute resources processing PII require runtime protection beyond traditional endpoint security. Containers inherit vulnerabilities from base images. Serverless functions access PII through API calls that bypass network perimeters. Deploy cloud-native application protection platforms that secure workloads from build time through runtime.
• Cross-environment visibility: Hybrid deployments split PII processing between on-premises and cloud systems. Unified monitoring across environments detects attacks that span infrastructure boundaries, enabling investigation of breaches involving data movement between cloud and on-premises systems.

Autonomous cloud workload security becomes essential when manual monitoring can't match the pace of dynamic cloud operations.

Protect PII Data with SentinelOne

Your PII security requires autonomous response that matches AI-optimized attack speed. Manual correlation and signature-based detection can't counter the high credential stuffing rate or deepfake social engineering targeting your systems right now.

Singularity Platform's behavioral AI detects the credential stuffing by identifying anomalous authentication patterns that signature-based systems miss, providing continuous monitoring across AI systems as required by NIST CSF 2.0. The platform executes NIST's monitoring requirement in real-time, spotting AI-enhanced attacks the moment they deviate from normal behavior patterns.

Storyline automatically creates forensic-quality investigation data with complete context linking attacker actions to specific PII records. You get the complete audit trails NIST Privacy Framework 1.1 demands without manual log correlation. When regulators investigate your breach, Storyline delivers evidence demonstrating exactly what your AI systems accessed, processed, and output: satisfying accountability requirements under GDPR Article 5(1)(a) and CCPA §1798.185.

Singularity Cloud Security includes integrated DSPM that automatically discovers, classifies, and prioritizes exposures within cloud object datastores and relational datastores in accordance to PII protection mandates such as GDPR, SOC2, and NIST 800-122. With continuous protection provided by DSPM, you will have an always-up-to-date understanding of where sensitive data related to SSNs, CCNs, HIPAA and more is stored and whether it’s exposed to attackers. 

Purple AI handles routine threat evaluations so your analysts can focus on complex investigations involving PII compromise, directly executing NIST CSF 2.0's RESPOND function requirements. The AI evaluates alerts autonomously, triages incidents by severity, and escalates only those requiring human expertise. Your team investigates PII breaches, not false positives.

Request a demo to see how SentinelOne secures PII data in the age of AI.

FAQs

What is PII?

PII stands for personally identifiable information. It encompasses data that identifies individuals either alone or combined. NIST defines PII as information distinguishing identity: name, SSN, biometrics.

GDPR expands this to online identifiers (IP addresses, cookies, device fingerprints). CCPA includes keystroke patterns, gait analysis, and AI-generated inferences about psychological characteristics. Apply the most restrictive definition based on jurisdiction.

What is Personally Identifiable Information (PII) Security?

PII security encompasses the technical controls, policies, and procedures that protect personally identifiable information from unauthorized access, disclosure, modification, or destruction. 

It includes encryption, access controls, monitoring, incident response, and compliance with regulatory frameworks governing data protection across collection, storage, processing, and disposal.

How to Detect and Respond to PII Data Breaches

Deploy continuous monitoring to detect anomalous access patterns, unauthorized data transfers, and system compromises. When breaches occur, immediately contain affected systems, assess the scope of compromised PII, notify affected individuals and regulators within required timeframes, and implement remediation measures to prevent recurrence.

How PII Is Collected, Stored, and Transmitted?

Organizations collect PII through web forms, applications, transactions, and automated systems. Storage occurs in databases, cloud platforms, and enterprise applications with encryption protecting data at rest. 

Transmission happens via encrypted channels during authentication, API calls, and inter-system communications, with access controls governing who can process data at each stage.

What are the Common Types of PII Data?

Common PII types include names, Social Security numbers, email addresses, phone numbers, physical addresses, dates of birth, financial account numbers, biometric data like fingerprints and facial recognition, medical records, IP addresses, device identifiers, and behavioral data including browsing history and location tracking.

Who is Responsible for Safeguarding PII?

Organizations collecting or processing PII hold primary responsibility for safeguarding data. This includes executives setting security strategy, security teams implementing technical controls, IT staff maintaining systems, employees handling data appropriately, and third-party vendors processing PII on your behalf through contractual agreements.

What role do employees play in maintaining PII security?

Employees serve as the first line of defense against social engineering, phishing, and insider threats. They must follow data handling policies, recognize suspicious activities, report security incidents promptly, maintain strong authentication practices, and understand their specific responsibilities for protecting PII within their roles.

How has AI specifically changed PII threats in cybersecurity?

AI created seven new attack methods: 19% credential stuffing rates through optimized authentication attempts, deepfake impersonation costing $25 million, data poisoning corrupting AI models, training data supply chain attacks, adaptive malware, AI reconnaissance mapping PII databases, and enhanced phishing. 

AI also expanded what constitutes PII: behavioral biometrics, voice prints, health data from wearables, and algorithmic inferences.

What frameworks should security teams use for AI PII protection?

Adopt an integrated approach combining NIST Privacy Framework 1.1, NIST Cybersecurity Framework 2.0, CISA AI Security Roadmap, and SANS Critical AI Security Guidelines, with compliance to GDPR, CCPA, and HIPAA based on jurisdiction. Integrate NIST Privacy Framework 1.1 with CSF 2.0. Layer CISA's Secure by Design principles and SANS guidelines.

Do AI-generated inferences count as personal information?

Yes, under California law. CCPA §1798.140(o)(1)(K) categorizes AI-generated inferences as personal information: "Inferences drawn from information to create profiles reflecting consumer preferences, characteristics, psychological trends, behavior, attitudes, intelligence, abilities, and aptitudes." 

Risk scores, behavioral predictions, psychological assessments, and preference predictions generated by AI constitute personal information requiring protection. GDPR Article 4(1) coverage creates similar obligations.