Every time I put a system in front of a company that decides on its own, the same moment arrives: someone in the room asks, "And if it gets it wrong, who is accountable?" It always comes with half a smile, as if this were a philosophical question that can wait for the next meeting.
It cannot. Israel has already answered it in writing, on two separate fronts.
The Regulator Already Answered: 100% On You
On 24 December 2025 an interministerial team published its final report on the use of artificial intelligence in Israel's financial sector. Its members came from the Ministry of Justice, the Ministry of Finance, the Banking Supervision Department, the Capital Market, Insurance and Savings Authority, the Israel Securities Authority and the Competition Authority.
The core ruling, as Calcalist reported it: responsibility for any damage an AI system causes a customer falls entirely on the supervised entity. Not on the algorithm's developer. Not on the outside software vendor. And you cannot cap that liability with a clause in the customer contract.
Beyond the ruling, the report sets four requirements that translate straight into work. Build an AI governance framework: policy, approval processes, control and monitoring. Hold a higher bar of control, explainability and human involvement in uses with material impact on customers, such as credit, insurance and investments. Disclose to the customer that an AI system is in use, calibrated to the level of risk, and offer a human service alternative. And run due diligence on vendors, with agreements that spell out transparency duties and cooperation in audits.
Not a supervised financial body? Do not relax yet. This is the most detailed policy statement Israel has produced on AI accountability, and when a regulator reaches your sector it will not start from a blank page. It will start here.
Amendment 13 Made It Personal
The second front is closer. Amendment 13 to the Privacy Protection Law took effect in August 2025. It sharply expanded the Privacy Protection Authority's enforcement powers, required certain organizations to appoint a privacy protection officer, and for the first time imposed direct liability on senior officers.
This is the point where I ask every CEO to stop. Before the amendment, a data protection failure was a corporate risk. After it, the risk is also personal, and it belongs to whoever sits in management and on the board.
Enforcement has already started. On 21 July 2026 the Authority imposed the first financial sanction under the amendment, NIS 256,000, on a healthcare organization, for failing to report a serious security incident immediately. The Authority stressed that the duty to report arises the moment the incident becomes known, without waiting for the checks to finish. According to a law.co.il report from late July 2026, the Authority is running more than 100 enforcement files.
Now set that beside one finding from the State Comptroller's report of June 2026: 41% of the bodies audited operate with no formal data governance framework, and only 18% have adopted an AI strategy or policy. The same picture repeats globally. In IBM's Cost of a Data Breach 2025, 63% of organizations reported no AI governance policy at all.
What an AI Governance Document Must Actually Contain
I admit that "AI governance" sounds like one more binder nobody will open. So change the framing: this is not a compliance document, it is your defense file. The day the Authority, the board or a plaintiff asks "what did you do", this document is the answer. If it does not exist, there is no answer.
In my experience a document that works runs six to ten pages, not sixty, and it has six parts:
Use mapping. A living list of every place AI touches a business process. Which tool, who owns it, what data goes in, what comes out. Without that list, everything else is theoretical.
Risk classification per use. Not every use carries the same weight. Drafting a marketing email and scoring a credit application are not the same animal. Three levels are enough, and what sets the level is the impact on a third person.
Human in the loop on high-risk decisions. Write it out: which decisions do not go live without a named human approval, and who is authorized to approve. That is exactly what the interministerial report demands for uses with material impact on customers.
Documentation. Logs of input, output and who approved. The regulator will not ask whether there was a mistake. It will ask whether you knew, when, and what you did about it.
Incident procedure. Who detects, who gets told, within what time. Note that the law has already set the bar: immediate reporting, not complete reporting.
Customer disclosure. When you tell a customer they are talking to a system, and how they ask for a person.
Five Questions for the Board
The Israel National Cyber Directorate, in a report published in May 2026, recommended a binding date, 31 May 2026, for two things: presenting the threat from the new generation of AI models to boards and audit committees, and challenging the organization's risk appetite. These are the five questions I suggest raising there:
- Which decisions in our organization are made today with AI, and which of them affect a customer or an employee?
- Who in management personally owns this, by name and title?
- What happens if our AI vendor goes down tomorrow morning, and how long will it take us to get back to manual work?
- If an incident surfaces tonight, who do we report to, within how many hours, and who writes the report?
- Where is the document that records the answers to the four previous questions, and when was it last updated?
If the answer to the fifth question is silence, that is where you start.
I am not a lawyer and I am not a regulator, and at the points where a legal opinion or a professional security audit is needed, I say so plainly and refer it out. What I do know how to do is sit with a CEO and map, inside a day, what is really running in his organization, so the document describes reality rather than intentions. If that is worth a conversation to you, I am here.
