Days before Israel's 27 October Knesset election, the country has a working rule for one kind of AI harm and nothing for another. The Central Elections Committee, chaired by Justice Noam Sohlberg, requires labels on AI-generated campaign material. Yet, as the Times of Israel reported on 19 October, the rules address deepfakes, not the way voters increasingly ask chatbots what to think.
What the rules actually require
The committee published its rules on 26 July 2026, under a recently amended provision of the Elections (Propaganda Methods) Law, according to a Pearl Cohen summary. Campaign material that is generated or substantially manipulated by digital means, and that could be mistaken for authentic content, must carry a prominent disclosure. That can be a text notice or one of two standardized symbols: one for content created wholly or mostly by digital means, the other for content materially altered.
The format rules are concrete. For printed material and images, the notice must be at least 5% of the content's height or 10 points, whichever is greater. Video notices must stay on screen throughout, and audio-only content must carry a spoken notice at the beginning and end (Israel National News). Attaching a compliant notice creates a presumption of compliance. The Times of Israel reports the rules stay in force until a year after the election.
The strongest case for this approach
The committee's defenders have a good argument. Election law is the one area where Israel already has an enforcement body, a statutory hook and a short timeline. A label is a narrow, speech-preserving remedy: it does not ban satire, ads or synthetic media, it only tells the audience what it is looking at. Acting through existing election law avoids the risk of rushing a sweeping AI statute through a campaign period. Labels are also a form of regulation that a free-speech publication can accept, because they add information rather than remove speech.
We agree with that. Mandatory disclosure on realistic synthetic political content is proportionate, and it is far preferable to takedown mandates.
Where the rules stop
The difficulty is that the labelling rule governs publications by campaigns, while the Times of Israel's experts say the more consequential channel is the chatbot. The Israel Internet Association's 2026 survey found 88% of Israeli adults use ChatGPT and 73% use Gemini, per the Times of Israel. Voters who ask a chatbot about a candidate receive an answer, not a campaign ad, and no label rule reaches it.
The article describes generative engine optimization (GEO) as the chatbot-era equivalent of search engine optimization: planting content designed to shape what AI systems retrieve and repeat. Tehilla Shwartz Altshuler of the Israel Democracy Institute is quoted predicting that GEO "will decide the elections." Shay Hershkovitz, a Georgetown adjunct professor, argues that synthetic video has so far underperformed fears about vote-flipping, while poisoned retrieval is harder for voters to detect because the chatbot states distorted material as a straight answer.
This matters because a deepfake is a discrete artifact that can be debunked. A skewed chatbot answer is private, personalized and unrecorded. A label on a campaign video does nothing about it.
No statute, no platform law
Israel has no AI statute. A Legal500 guide states that no binding AI regulations have been formally enacted, although bills, including on deepfakes, have been proposed. The government's December 2023 policy on AI regulation and ethics, published on gov.il, chose a "Responsible Innovation" model. As Legal500 summarizes it, the model favours sector-specific regulation by existing regulators, a risk-based approach aligned with OECD principles, international interoperability and incremental, soft-law tools.
That is a defensible design for a small, innovation-heavy economy, and in ordinary times we would defend it. But it assumes a sectoral regulator exists for each risk. For election-time chatbot influence there is none, and the Times of Israel quotes Asaf Wiener of the Israel Internet Association saying Israel has no meaningful regulation or accountability for platforms. Experts cited there point to the tech sector's economic weight and US pressure as reasons for reluctance.
The EU offers a contrast, though not necessarily a model. The AI Act imposes transparency duties on AI providers and deployers, and the Digital Services Act requires very large platforms to assess systemic risks, including to electoral processes. Those frameworks carry real compliance costs and the DSA in particular invites regulators to judge lawful speech. Israel does not need to copy them.
A proportionate path
There are narrow options consistent with Israel's own policy. Each avoids content mandates and none requires a new AI law:
- Transparency, not control. Ask the largest chatbot providers to publish how they handle election queries, including source selection, and to disclose changes during campaign periods.
- Measurement before mandates. Fund independent audits of chatbot answers about parties and candidates, so claims about GEO manipulation rest on evidence rather than prediction.
- Literacy. Experts in the Times of Israel report describe digital and media literacy as very low relative to platform adoption. That is a gap education can address without restricting anyone's speech.
- Sunset and review. The committee's rules expire a year after the election. That is a sensible habit: the Knesset should use the window to review what labelling achieved before legislating.
What to watch
The test of the labelling regime will be modest: whether campaigns comply, and whether the committee enforces. The harder test is whether the next Knesset treats chatbots as an election-integrity question at all. A policy built on sectoral regulators only works if someone is assigned the sector. At present, nobody is assigned this one.