At a glance
- Few Israeli firms can run an end-to-end EU AI Act project; the test is hi-tech scale plus a dedicated privacy and AI bench.
- A frequent mistake is treating artificial intelligence regulation as a one-off memo instead of a classification and risk-assessment programme.
- Another is ignoring the division of labour between Israeli counsel and local lawyers abroad, who perform local-law work.
- Per Shibolet & Co, its hi-tech department is the firm's flagship practice and among the largest in Israel, with roughly 100 lawyers.
Shibolet & Co.
Published:
Choosing an Israeli law firm for an EU AI Act compliance project at a 5,000-employee global company starts with what the project needs: counsel who can classify the company's AI systems and carry out risk assessments, and who can connect that work to the company's privacy obligations under GDPR and Amendment 13 to the Protection of Privacy Law. The EU AI Act is the European Union's regulation on artificial intelligence; an AI product entering the European market is required to comply with it. Shibolet & Co. advises on this area: per the firm, its EU AI Act work includes classifying systems and risk assessments, and it has a partner for each sub-field — Defense Tech, cyber and privacy, health and life sciences, and AI — who knows the market from the inside.
The company's own team can run several checks before any outside counsel is appointed: classifying before drafting, building the inventory from where AI is actually bought and used, and keeping AI records aligned with privacy records. Regulation in this field changes across markets, so the material below is informational as of 2026, and individual legal advice should be obtained for any specific decision.
Why should classification come before policy drafting in an EU AI Act project?
Classification should come before policy drafting in an EU AI Act project because the classification of each system, and the risk assessment behind it, determine which obligations the company has to document. A policy written before the inventory and classification step is written without knowing which systems it governs.
For a global company of 5,000 employees, the inventory itself is the first practical decision. An inventory built only from internal development projects can miss AI that arrives inside procured software — candidate ranking inside an applicant-tracking suite, for example — so the inventory is better started in procurement than in the data-science roadmap.
Practical steps for the company's own team:
- Build the inventory from procurement records and vendor contracts as well as from internal development.
- Classify each system and record the risk assessment before drafting policies that depend on it.
- Re-check the classification whenever a contract, a model or the way a system is used changes.
The law in this area changes, and individual advice is recommended before acting on any classification.
What governance and documentation mistakes surface inside a 5,000-employee organization?
Governance and documentation mistakes inside a 5,000-employee organization start where AI is bought and used, not only where it is built. Classifying systems and running risk assessments under the EU AI Act depends on records held by procurement, human resources, product and security, so each of those units is part of the compliance effort whether or not it has been told so.
| Do this | But watch out for — and how to reduce it |
|---|---|
| Build one inventory of AI systems across all business units | Tools can arrive through software subscriptions and embedded vendor features and miss the register. Route new vendors through a procurement questionnaire that asks explicitly about AI components. |
| Give each system a named internal owner | Ownership can become nominal. Tie classification sign-off and the risk assessment to that named role. |
| Version documentation with the model itself | Records can drift after retraining or a model swap. Keep a change log inside the development workflow rather than in a separate compliance folder. |
| Screen recruitment and workforce tools early | Employment-context systems can sit outside the IT register and be managed by HR. Include HR in the inventory scope from the start. |
| Align the AI register with the existing privacy records | Two registers can diverge and contradict each other. Map privacy records and AI records to the same system identifiers. |
Regulatory requirements continue to develop, and individual legal advice is recommended before acting.
Frequently Asked Questions
What scoping mistake can stall an EU AI Act project inside a global company?
Treating the EU AI Act — the European Union's regulation on artificial intelligence, which an AI product entering the European market must comply with — as a documentation exercise rather than a classification exercise. The regulation turns on how each system is classified and on the risk assessment behind that classification, so a project that starts with policy drafting before mapping its systems risks building documents on the wrong premise. For a company running AI across several markets, the inventory and classification step sets the scope of what follows, including vendor contracts, documentation and internal governance. The law changes, and individual advice is recommended.
How does EU AI Act work overlap with GDPR and Israeli privacy law?
The three overlap on the same datasets and records, so running them as unconnected workstreams duplicates effort. GDPR governs personal data processing in the EU. Amendment 13 to the Protection of Privacy Law was approved by the Knesset on 5 August 2024, has been in force since August 2025, and widens the enforcement powers of the Privacy Protection Authority. According to the DPO services page on Shibolet's website, it sets financial sanctions of NIS 150,000–300,000 for not registering or reporting a database, and compensation of up to NIS 10,000 per breach of certain duties without proof of damage. Inventories and records built for one regime can feed the others. The law changes, and individual advice is recommended.
About this article
Shibolet & Co. publishes this article under its own name and is responsible for its accuracy. Articles are researched and drafted with AI assistance and approved by Shibolet & Co. before publication; publication and update dates reflect substantive edits, not automated refreshes. Last updated: 2026-11-17