Deepfake and Election Campaign Law

The petition concerned an image published by the Likud party on social media, depicting Naftali Bennett and Yair Lapid in a manipulated situation (photomontage or AI-generated content), creating a false impression of political cooperation with Arab parties or agreements that never took place in reality. A temporary injunction was sought to remove the publication and compel the party to publish a clarification that the image had been doctored.

The Legal Definition of Deepfake

The dangers inherent in such uses, as well as a definition of the term, were established in a similar case several years ago – in the context of a petition for an injunction under Section 17b of the Elections Law (Propaganda Methods) 5719-1959, filed by Yesh Atid, led by Yair Lapid, against the “Yes to Peace” association and Facebook Israel, seeking the removal of a deepfake video published on the respondent’s Facebook page. The petitioner argued that the video featured a figure purporting to be Yair Lapid delivering an “election speech.” In his decision, Supreme Court Justice Uzi Vogelman addressed the term:

“Deepfake is a term for technology used to create audio or visual content, or to alter existing content, such that the reasonable viewer (and even the sophisticated viewer) would believe that a certain person performed an action or conveyed a message, when in fact the content is not real. The content is of such high quality that the average user would typically struggle to detect that it is a fabrication.” (Election Campaign Case 24/9, 2021).

The Legal Question

The question before the court was whether the publication of a realistically manipulated image (Deepfake/AI) constitutes a violation of Section 13 of the Elections Law (Propaganda Methods), which prohibits the publication of election propaganda containing “public disturbance” or “false information” capable of influencing a voter’s will in a misleading manner:

“Election propaganda on behalf of, or in favor of, one party or list of candidates shall not be conducted in a form or manner that constitutes an unfair interference with election propaganda on behalf of, or in favor of, another party or list of candidates.”

Justice Sohlberg’s Position

Justice Sohlberg noted that beyond speculation about the use of artificial intelligence in this case, no evidence whatsoever had been presented to that effect – no expert opinion, and no features unique to artificial intelligence in the manipulated image. This was noted alongside a reference to remarks he made in a previous petition that was dismissed for failure to give prior notice to the other side: “It is unclear whether the image at the center of this petition was created using artificial intelligence, or perhaps using ‘traditional’ computing tools such as standard image-editing software” (paragraph 3).

Nevertheless, Justice Sohlberg found it appropriate to make two observations applicable to “engineered” publications – whether created using artificial intelligence or other means:

First Observation – The Danger of False Publications

Publications of this kind risk undermining public confidence in the integrity of elections altogether. As Justice Vogelman put it: “If everything can be faked, why believe anything at all? It follows that they pose a danger to the very existence of the electoral process and the functioning of the democratic system.” (Yes to Peace case, paragraph 8).

Second Observation – Violation of Section 13

In this case, a violation of Section 13 was found, regardless of the means used to create the image, in part because the publication appeared to be authentic. Had it been clearly and visibly indicated that the image does not represent reality and that it was produced by software, it is quite possible that a near-certainty of actual voter deception would not have arisen.

The Outcome

Justice Sohlberg clarified that protecting the integrity of elections requires combating the deepfake phenomenon, as political free speech does not include the “freedom to fabricate reality” using advanced technological means – and the petition was granted.

The key orders issued were as follows:

• Removal of the image – Likud was ordered to remove the image from all platforms on which it had been published in its current form.
• Publication of a correction – denied – The court declined to order the respondent to publish a statement acknowledging that the image had been doctored, due to the one-week delay in filing the petition and concern that such a publication would merely “resurrect” the issue and serve the propaganda a second time.
• Mandatory AI content labeling – deferred – Although the court acknowledged that the question of labeling edited publications is an important one, extending beyond election law, Justice Sohlberg refrained from ordering a clear labeling requirement for content edited or created using AI, ruling that “it is appropriate to give the legislature the opportunity to regulate this matter before a judicial determination is made.”

Click to read the full decision (In Hebrew)