AI tools for an Israeli business: what to check before they touch your work

The major tools work from Israel, handle Hebrew reasonably well, and can be running in five minutes with a credit card. That is why the question people ask is which one to pick, and it is not the question that costs anything if you get it wrong.

The question that matters is what information may go into the tool, and who answers for it if that information gets out. AI did not change that rule, but it did make it practical rather than theoretical: until recently nobody pasted their customer list into a foreign company's website to tidy up a table.

What changed in the law

In Israel, Amendment 13 to the Protection of Privacy Law came into force on 14 August 2025. It widened what counts as sensitive information, widened the obligations of anyone holding information about people, and gave the Privacy Protection Authority real enforcement powers. The penalties it carries reach millions of shekels, with multipliers for large databases and for sensitive data.

There are also obligations that begin above certain thresholds, such as appointing a privacy officer. Those thresholds depend on the kind of business, the volume of data and what is done with it, and they are not something a guide can settle on your behalf. If you hold customer data at any non-trivial scale, that is a question for a lawyer rather than for a search engine.

Why this bears on AI tools in particular

An AI tool is a third party. The moment you paste information about people into it, you have moved personal data to another organisation, usually outside Israel, without an agreement saying what it does with it. That the tool is available and speaks Hebrew does not make that particular use of it sound.

And here is the part that trips people up: people separate "a system" from "a chat". The law does not. A chat box is a transfer of information exactly as an import into a file is.

What to check before a tool touches your work

• Whether there is a business agreement making the vendor a processor of your data, rather than only consumer terms of use.
• What is retained, for how long, and whether your input is used for training.
• Who on the team can sign in, and what happens to the access of somebody who leaves.
• Whether what you plan to put in includes information about identifiable people. If it does, stop here.

And on Hebrew

These tools are weaker in Hebrew than in English, and it does not show up as obvious errors so much as phrasing that looks correct and does not sound like a person. On this site we hit the extreme version: generating Hebrew text inside images is so unreliable that our art-direction brief forbids text in images altogether.

The practical conclusion is not to avoid them, but to test on your own text rather than on a demo. Five minutes on something real of yours is worth more than any comparison you will read, this one included.

The full list of what does not go into an outside tool is in what not to paste into an AI tool.

A note: this is an explanation, not legal advice. What is right for your business depends on your data and what you do with it, and that is a question for a lawyer.

Build a system that processes real data in a defined way