Israel's Securities Authority has put out for public comment a draft directive that would set risk-tiered technology and AI duties for investment advisers, marketers and portfolio managers

What did the Securities Authority publish, and who would it reach?

On 4 November 2026 the Israel Securities Authority released a draft directive for public comment. Its subject is how licensees under the Investment Advice Law use technology, and AI in particular, when they give investment advice, market investments or manage portfolios. The cover marks it as a text for public comment, with comments accepted until 6 December 2026.

It is a draft. Nothing in it applies to anyone today. The Authority calls it a new version of its existing directive, which dates from 2016 and has been updated several times. The explanatory notes say the proposal keeps the existing duties and keeps the older text's technology-neutral approach, so that duties follow the nature of the service rather than the type of tool.

The background section traces the draft to an inter-ministerial team set up at the end of 2022, whose final recommendations came out in December 2025 in a report titled Artificial Intelligence in the Financial Sector. For investment advice, marketing and portfolio management, that team recommended adding a dedicated AI chapter to the existing directive. This draft is the Authority's version of that step.

How would the draft sort AI use by risk?

The draft adopts the team report's definition of an AI system: a computer-based system operating with varying degrees of autonomy to produce outputs such as content, predictions, recommendations or decisions that may affect investors, clients or the supervised body's activity. A licensee may decide that a system falls within the wording but is not AI by its nature. The notes add that this is not an automatic or blanket exemption and has to rest on a case-by-case look at the system, its purpose and its risks.

Before using a technological means, a licensee would assess the risk of the tool and the service delivered through it, and separately assess its materiality. Draft clause 8 lists the minimum factors: how far the licensee relies on the tool, the scale of use, the effect of a failure on the service, the degree of human involvement including whether the service runs without active human involvement, and the way and purpose of use. The assessment would be revalidated at least once a year or on a material change, and documented.

On our reading, three tiers result. Online-only service, meaning a tool used merely as a communication channel and not for decisions, carries a reduced set of provisions. Ordinary technology use carries the general duties. And 'heightened use', defined as a service using a technological means, other than online-only service, that the licensee has itself classified as high in risk and materiality, carries more.

Heightened use would be delivered only by a corporation and its licensees, not by a sole practitioner (draft clause 19). It would also have to be reported to the Securities Authority when it starts or ends, along with any significant change that amounts to a material event, with the corporation spelling out why its activity qualifies (clause 20).

What governance and monitoring duties does the draft propose?

The draft would have a licensee using a technological means set written policies and procedures for managing the risks (clause 13) and keep those procedures, and superseded versions, for seven years (clause 14). Clause 5 limits online-only providers to clauses 1 to 6, 9, 13 to 15, 23, 25 and 26, 29, 31 to 33, 35 to 40, and 43 and 44, so the technology officer and staffing requirements that follow apply to other licensees. The draft would have the licensee appoint a dedicated technology officer, who need not hold a licence and whose appointment does not shift responsibility away from the licensee (clauses 16 and 17). Clause 18 would have the licensee employ additional staff, and a corporation additional licensees, in a scope that lets it meet the directive.

For heightened use, the board of the corporation would be responsible for approving the technology-use strategy including outsourcing, the risk and materiality assessment and its results, the tool itself and its use (with the approval record kept for seven years from the last service given with it), the procedures, and the technology officer's appointment (clause 21). The corporation would also run periodic training for its board (clause 22).

Monitoring is its own chapter. Outside that online-only list, a licensee would operate monitoring and control to check that input data is of good quality and reliable, that outputs are too, and that outputs are not discriminatory, biased or driven by potential conflicts of interest. It would identify and handle abnormal outputs, including the option to stop or limit the tool, and have alerts for material deviations in the tool or model performance (clause 24). It would keep a register of all technological means in use (clause 25) and retrievable documentation of the service process for seven years (clause 26).

The explanatory notes give examples of what the Authority has in mind on security: input data and how it is processed and stored, protection against leakage or misuse, and cyber risks such as prompt injection and data poisoning. Those examples sit in the notes, not in the operative clause.

What would the draft require on explainability, disclosure and outsourcing?

Explainability comes in two levels. Outside the online-only exemption in clause 5, a licensee using a technological means would need to understand it well enough to explain its logic, key limits, the data used and the considerations behind its decisions (clause 27). For heightened use, the licensee would also need the ability to explain why a particular output was produced rather than another, but would not owe that where compensating measures satisfy it that the absence is covered. Those measures have to reduce the resulting risk significantly (clause 28). The explanatory notes add that where compensating measures do not reduce the risk enough, the licensee would not be permitted to provide the service using that tool.

On disclosure, a licensee would give clients, before the service starts, a disclosure of the service's distinct characteristics compared with a service not given through technology, and the risks (clause 29). Outside the online-only list, it would show a clear, prominent and effective indication that the service includes an AI system, except where that is already obvious (clause 30). The draft also bars misleading statements about what a tool can do and, in the notes, ties this to AI washing in advertising (clause 31).

Outsourcing gets a dedicated chapter, clause 34, which is also outside the online-only list. The licensee would build outsourcing into its policy, assess the particular risks and run due diligence on the provider before contracting, keep the knowledge and permissions it needs to monitor effectively, and make sure the contract covers incident reporting, data protection, sub-contractors, material changes to the tool or underlying model, data return and deletion, exit, and service levels for faults. It would set business-continuity principles and disclose to clients before contracting if a provider exit would likely stop continuous service. The licensee's accountability and the Authority's access to information would be preserved.

Two service-stage items are worth a separate note. A licensee could decide whether to offer human assistance at all, provided it says so clearly before the client signs on, though anything the law reserves for a licensee would still be handled by one (clause 37 and the explanatory notes). And a portfolio run by an algorithm or AI at a licensed corporation connected to a banking corporation would contain only units of mutual funds, including foreign funds the Authority has allowed to be offered in Israel (clause 42).

What is the Authority asking the public, and what is still open?

Beyond the draft text, the Authority's staff asks for views on four things. First, whether to set uniform quantitative or qualitative criteria for heightened use rather than leave classification to the licensee. As an example the document points to the thresholds used to classify a large portfolio management company under the Investment Advice Law: more than 50 clients and total assets above NIS 5 billion, or more than 1,000 clients. Second, whether to add dedicated requirements for the board and management of a corporation that uses technology heavily, such as a minimum number of directors, board expertise or committees.

Third, whether the use of AI agents by financial bodies raises unique risks, or magnifies existing ones, that the draft does not cover, and if so which. The document describes agents as systems that perform complex tasks autonomously or semi-autonomously, planning action sequences, using external tools and systems, and sometimes with limited human involvement.

Fourth, the connection of general AI models to financial platforms: a client linking financial data held at a financial body to an outside generative AI tool or agent, sometimes from the body's own interface and at its encouragement. The Authority lists questions it is examining, including whether this amounts to unlawful referral of clients to investment advice under section 3A of the Investment Advice Law, whether it can create a false impression that the outside tool's activity is a supervised service, and what privacy and security issues arise when sensitive portfolio data goes to tools outside supervision. It states that raising the topic is not an explicit or implied permission to carry out the activity.

None of these four questions has an answer yet. What happens after the comment period, including any changes to the text, is not described in the document.

What we did not verify

What we opened: the Israeli Securities Authority's draft directive itself, a Hebrew PDF, fetched directly from the URL in the sweep and extracted with pdftotext in UTF-8. The cover page, explanatory notes, the 44 draft clauses and the closing section on additional questions were all read in the Hebrew. We confirmed the document's own title, the 4 November 2026 date, the 6 December 2026 comment deadline, and clause 43 on the start date.

What we did not open: the inter-ministerial team's December 2025 report, the existing 2016 directive, the Investment Advice Law, the Authority's general guidance on how it sets regulation, and any English-language summary. Our description of those comes only from what this draft says about them. We did not locate any press release, comment-intake page or consultation timetable beyond the cover page.

What we refuse to claim: that this text will be adopted as drafted, when it would start, how many licensees would count as heightened users, or how the Authority would enforce it. Clause references are to the numbering printed in the draft and could change. Quoted Hebrew is limited to the pull quote; everything else is our paraphrase and any English wording of a clause is ours, not an official translation.

Informational analysis for working professionals, not legal advice. Confirm how any rule applies to your situation with qualified counsel.

Related briefings

• Israel PPA Amendment 13 and AI enforcement
• Bank of Israel Directive 369 on AI model risk
• IOSCO AI supervisory toolkit
• Middle East and Africa AI regulation tracker
• AI Regulation News hub