שפת ממשק
רגולציהתקנות אבטחת מידעcontent.hero.readingMinutescontent.hero.updated May 14, 2026

Israel's Privacy Security Regulations and AI — What Regulation 13 Requires

Israel's Privacy Protection Regulations (Information Security) 5777-2017 are the practical foundation for data security requirements on personal databases in Israel. Regulation 13 imposes heightened obligations on high-security databases — comprehensive documentation, periodic audits, and structured access rights management. When an organization processes personal information using AI, these regulations apply in full to every database the AI system touches.

Year enacted
2017
Security levels
5
Audit log retention
24m
Internal audit cycle
2y
content.layout.tocMobileLabel
  1. 01 Five security levels
  2. 02 Core obligations
  3. 03 Five immediate actions

Overview — Five Database Security Levels

The Privacy Protection (Information Security) Regulations classify every database into one of five security levels:

  • Basic database — Minimum requirements (Regulation 4)
  • Medium-level database — Requires a documented security procedure (Regulation 5)
  • High-level database — Heightened requirements, including annual reviews (Regulation 13)
  • Unique-scope database — More than 5 million data subjects
  • Special-type database — Databases held by certain public bodies

Most SaaS, fintech, and healthcare organizations in Israel are classified as high-level databases — which is precisely when Regulation 13 comes into force. For a complete map of database types, see Israeli Database Law and AI.

Core Obligations Under Regulation 13

1. Access Rights Management

Organizations must document who is authorized to access the database, at what permission level (read / write / delete), and on what basis the authorization was granted. In the AI context this means answering:

  • Which employees may feed data from the database into an external AI tool?
  • Which AI systems hold read access to the database API (for example, a chatbot pulling customer records)?
  • Which users may download model outputs that contain personal information?

This list must be updated whenever a role changes, and access must be revoked immediately upon termination of employment.

2. Access Activity Logging (Audit Log)

Every access to a high-level database must be recorded in an immutable log. In the AI context, this requirement has significant practical implications:

  • Logging every prompt that includes personal information
  • Logging every API call from an AI system to the organizational database
  • Logging which data was exported to an external AI tool, when, and by whom

3. Periodic Access Review

Regulation 13 requires a periodic review of authorized users — typically at least once a year. For AI systems, this review must also cover: which service accounts and API keys have access to the database; whether external vendors are still active; and whether DPA agreements with those vendors remain current.

4. Information Security Incident Management

Regulation 13 imposes an obligation to document and handle every information security incident — including unauthorized access attempts, inadvertent exposure, and data loss. AI introduces a new category of incidents that organizations must explicitly define:

  • A model "hallucination" that exposed one user's private data to another
  • An employee feeding sensitive data into an external AI tool without a DPA in place (unauthorized use)
  • A vendor-side model update that degraded output quality or introduced bias

A serious incident (affecting 10,000 or more data subjects, or involving specially sensitive information) must be reported to the Authority within 72 hours of detection. See Amendment 13 to the Privacy Protection Law for the full notification obligation.

5. Periodic Internal Audit

A high-security database must undergo an internal audit at least once every two years. Forward-looking organizations align this audit with their ISO 27001 and ISO 42001 review cycles — significantly reducing duplicated effort. See our guide to integrating the two standards.

Five Immediate Steps for Regulation 13 Compliance in an AI Context

  1. 1

    Map authorized access

    Within 30 days: identify every employee who can feed database data into an external AI tool. Document the list and revoke any unnecessary access.

  2. 2

    Update DPA agreements

    Ensure every AI vendor that processes data from the database has a written contract with information security clauses appropriate to the database security level.

  3. 3

    Enable audit logging

    Activate logging on every API access to the database, including access from AI systems. Retain logs for at least 24 months on tamper-evident storage.

  4. 4

    Add AI-specific incident procedures

    Extend your existing incident response procedure to cover AI-specific events: model hallucinations that expose personal data, unauthorized employee use of external AI tools.

  5. 5

    Plan a combined annual audit

    Schedule the next internal audit to include AI controls. The portal generates a specific checklist of AI control points aligned with Regulation 13 requirements.

content.layout.faqHeading

content.layout.faqCount
Do the Information Security Regulations apply to an AI system?

Yes, as soon as the system processes personal information that forms part of a registered database. If an organization feeds customer data into an external AI tool (ChatGPT, Claude, Copilot) for processing, that constitutes a 'transfer to an external party' and falls squarely within the regulations.

What is a 'high-security database'?

Regulation 1 defines a high-security database as one that processes information on 100,000 or more individuals, or whose purpose 'involves special sensitivity' — medical information, financial condition data, location data, or information about minors.

What must be documented under Regulation 13?

A database definition document, an information security procedure, an authorized-access list, a list of external vendors, a security-incident register, and periodic internal audit reports (at least once every two years). In the AI context: also a list of AI systems accessing the database, and DPA agreements with all AI vendors.

Is a tool like ChatGPT a 'holder' of the database?

Most likely yes, if the organization feeds personal information into it. The law distinguishes between the 'database owner' (the organization) and the 'holder' (an external party processing on the owner's behalf). This requires a written contract with appropriate DPA clauses and a review of the vendor's security posture.

What are the penalties?

Until 2024, penalties were relatively modest. Amendment 13 to the Privacy Protection Law (in force from 2026) raises them significantly — up to 5% of annual turnover or ILS 3.2 million. See the article on Amendment 13 for details.

content.layout.ctaBadge

content.layout.ctaDefaultTitle

content.layout.ctaDefaultSubtitle

content.layout.contactTitle

content.layout.contactBadge

content.layout.contactBody

content.topicsMap.eyebrow

content.topicsMap.headingTemplate

content.topicsMap.legendHint

content.layout.clusterLabel רגולציה

Israel Privacy Security Regulations — Regulation 13 and AI Compliance | Alice GRC Portal