Israel has no comprehensive AI act, and its election authority has just shown why it may not need one. On Sunday, July 26, 2026, rules issued by Justice Noam Sohlberg, who chairs the Central Elections Committee, took effect. They require prominent labels on campaign material created or significantly altered with AI, and they apply to anything first published on or after that date. With the Knesset election set for October 27, 2026, this is the first real test of targeted, disclosure-based AI rules in a country that has otherwise stayed with soft law.
What the rule actually requires
The rule is narrower than the headline suggests. According to reporting on the committee's announcement, a label is required only where content depicting a person, place, event, document or object was created or materially altered in a way that could lead viewers to believe it is authentic documentation. Not every use of AI triggers it. Video labels must stay visible for as long as the altered content appears. Audio-only material needs an audible disclosure at the beginning and end, and cannot rely on a visual mark. Printed material and images must carry a label at least 5% of the content's height, according to the committee's announcement. (The Yeshiva World)
The design follows a proposal the committee's acting director-general, Din Livneh, presented to the Knesset Constitution, Law and Justice Committee in June. As the Times of Israel reported, the measure would not prohibit AI in political messaging. It is meant to increase transparency. The proposal cited penalties of up to NIS 30,000 or six months' imprisonment. (Times of Israel)
The strongest case for the rule
The case for acting is not hypothetical. Sceptics of light-touch regulation argue that voters cannot reliably tell synthetic footage from real footage, that a fabricated clip can shape opinion before anyone corrects it, and that elections are the one setting where a deception cannot be undone afterwards. Israel's campaign has supplied examples. The Times of Israel noted that the Likud party was ordered in February to remove a doctored image showing political figures in fabricated scenarios. On July 26, Sohlberg ordered Religious Zionism to remove a video showing an AI-generated figure resembling Gadi Eisenkot shaking hands with Palestinian Authority Chairman Mahmoud Abbas. According to Israel National News, he found that it depicted events that never occurred and could falsely lead viewers to believe they were real. (Israel National News) In August the committee also ordered the Eisenkot campaign to remove an AI video of IDF soldiers, on the separate ground that it improperly used public property for election propaganda. (Times of Israel)
Why disclosure is the right instrument
Given that case, a label requirement is close to the least restrictive tool available. It leaves satire, illustration and obviously stylised content alone, because the trigger is plausible authenticity rather than the use of AI. It also sits within an existing regime. Israel's December 2023 policy on AI regulation and ethics, from the Ministry of Innovation, Science and Technology, favours sector-specific, risk-based, incremental rules over a single horizontal statute. (Israel Ministry of Innovation, Science and Technology) An election-period rule enforced by an election body is that model applied in practice. A regulator that already understands campaign law is better placed to define "deceptive" than a general AI authority would be.
The approach also fits a pro-speech position. A ban on synthetic political content would sweep in parody, commentary and legitimate campaign creativity, and it would hand incumbents a weapon against challengers. A rule that says "show your work" does neither.
Where the design is under strain
The first tension is that labels are not always treated as sufficient. According to Israel National News, Sohlberg rejected the argument that the small AI-edited caption on the Eisenkot-Abbas video was adequate, reasoning that Facebook users typically miss such labels. He ordered the video removed all the same. If a realistic fabrication can be taken down even when labelled, then the label requirement is not the operative constraint. Underlying prohibitions on unfair election interference are. That may be defensible, but campaigners deserve a clear line between what is permitted with a label and what is barred outright. Unclear boundaries chill lawful speech more than any label does.
The second is speed. A Jerusalem Post editorial argued that "a ruling delivered weeks after a fabricated recording reaches hundreds of thousands of voters is meaningless." It also noted that labels cannot stop viral spread across Telegram, TikTok, X, Instagram and WhatsApp, and that parties must answer for coordinated content distributed on their behalf. (Jerusalem Post) Those are fair points. Removal orders within days, as in the July 26 case, show the committee can move quickly. But the rule will only be credible if that speed holds across a campaign with many actors and platforms.
The third is the limit of the rule's reach. It governs campaign material, so it does not touch anonymous accounts or foreign influence operations. The same editorial noted that Israeli security officials have identified Turkey and Iran as foreign actors seeking to influence the vote, and that Israel lacks a comprehensive national policy and designated authority to counter such operations.
What to watch
The Central Elections Committee has been building capacity for months. It set up a dedicated team with the Shin Bet in January to examine AI's effect on the election, and in May it convened academics at Tel Aviv University on election integrity in the AI era. (Central Elections Committee) Three questions will decide whether the rule is a model or a cautionary tale. Does the committee publish consistent, reasoned decisions that distinguish labelled-and-lawful from unlawful? Does enforcement stay fast enough to matter? And does the rule stay confined to authentic-looking deception, rather than expanding to cover ordinary campaign editing?
If the answers are yes, Israel will have shown that a narrow, transparency-first rule can protect election trust without an AI act. If not, the label will be a formality and the burden will fall on takedown orders that are harder to defend on free-speech grounds. Other jurisdictions weighing election-specific AI rules should watch how the committee handles that gap between labels and removals.