Antisemitism in a Bikini

From Anne Frank to AI deepfakes, what is to be done about technology amplifying sexualised antisemitic memes?

Recent news concerning the generating and sharing on social media, specifically X, of sexualised images of women and children produced by Elon Musk’s AI known as Grok, has rightly caused an outcry. But what’s been left out of this furore is how such distasteful imagery is also directed against Jewish women for antisemitic attack.

Sexualised antisemitic imagery is found most notoriously in depictions of Anne Frank. Her AI-generated aged likeness appears in bikinis and in other alluring forms – eroticising images of a murdered child speaks, more than anything, to their creator’s dysfunctional impulses.

Yet Frank isn’t alone. Bella Wallersteiner, a public affairs executive and descendant of Holocaust survivors, was digitally stripped and placed in a bikini outside Auschwitz – other images depict Jewish women in swastika-decorated bikinis. Wallersteiner urged the UK government to act, prompting technology secretary Liz Kendall to condemn the images as “absolutely appalling”.

This phenomenon isn’t simply about AI though. In 2016, The New York Times was accused of invoking the historical trope of la belle juive (the “beautiful Jewess,” which casts Jewish women as inherently exotic and corrupting) by illustrating an article on Hebrew and the Sabbath with a stock photo of Natalie Portman in a bikini. Nevertheless, AI has intensified the problem. Scarlett Johansson has spoken of her likeness being used in deepfake porn, including a video viewed over 1.5 million times. And after October 7, Israeli women are routinely objectified. The “Israel Diaper Force” campaign depicts IDF women as diaper-wearing babies. Streamer Hasan Piker, with nearly three million followers, stated about the women who suffered that day that it “doesn’t matter if rapes f***ing happened”.

This AI-fuelled onslaught attempts to bully and silence Jewish and Israeli women online, in a digital economy where outrage, humiliation, even hate, generate profits through clicks. AI itself, however, is neutral, capable of helping not just harming. Israeli organisation AI for Good, for example, creates AI avatars of physically and sexually abused women. These avatars allow the women to tell their stories anonymously and safely.

Nonetheless, centuries-old antisemitic fantasies fusing sexuality, humiliation and violence, have returned to feed imagery that tips into something dangerous – as propogandists who attack Jews have realised. And Jewish women can experience a level of vulnerability and abuse over and above women generally who suffer sexualised image hijacking. Consequently, if platforms accept sexualised antisemitism isn’t simply a matter of obscenity or consent but of power, several things can be done.

AI systems don’t only need limitations on their ability to generate sexualised images of people without consent, but also on their ability to place them in abusive imagined contexts – as with the image of Wallersteiner in a bikini outside Auschwitz. This isn’t about restricting free speech but about product safety. Tools that cause predictable harm should not be allowed to operate unchecked. And sexualised content crossing the line into antisemitism, platforms should treat as aggravated abuse, not simply “offensive content”.

Indeed, here lies a problem with current content moderation approaches – whether by human or AI. These frequently focus on detecting offensive, abusive, of racially-hateful content. But antisemitism embodies many facets, including the historical – content moderation as it stands cannot therefore pick up the implicit Jew-hate in an aged rendering of a bikini-clad Anne Frank. At CAAI we’ve looked at the problem extensively – including by building a prototype AI-based tool that determines whether text or images are antisemitic. And we find that for content moderation to detect Jew-hate effectively it must incorporate dedicated classifications like IHRA (International Holocaust Remembrance Alliance definitions of antisemitism). The fact that many content moderators are based in Turkey and Pakistan, countries hardly friendly to Israel and Jews, is an associated problem that needs tackling.

Regulation, too, must catch up. Most existing frameworks still treat non-consensual sexual imagery and hate speech as separate issues, failing to recognise how often they intersect. Sexualised antisemitism sits precisely at that junction – as intersectional harm. Effective regulation needs drafting with this overlap in mind, informed by expertise from specialist Jewish organisations, women’s rights advocates, and legal scholars, rather than left to generic, reactive rule-making, or platforms themselves to self-regulate.

These policies must also be enforceable: with rapid takedown obligations, meaningful penalties for non-compliance, and transparency about how decisions are made. Without rigorous implementation and oversight, regulation risks becoming little more than a statement of intent while the intersectional harm and abuse continue unimpeded. Laws governing online safety – both locally (e.g. the UK’s Online Safety Act) and internationally – need to reflect this reality, or they will continue to miss the point.

There is also a cultural reluctance – often found in progressive spaces – to name antisemitism when it appears in sexualised or supposedly “humorous” forms. Yet with the degrading of Jewish women through images featuring erotic fantasy or infantilisation far from being a new phenomenon, at the very least, there must be historical awareness, so that these images are not dismissed as isolated incidents or edgy jokes but recognised as patterned abuse.

Latest events suggest X will stop Grok producing images of real people undressed after the backlash it received. Certainly Grok, and similar tools, need these stronger guardrails. But more than image generation it’s user distribution that’s the problem; and where sharing sexualised antisemitic content needs restraining – not least for Jewish figures no longer alive, like Anne Frank. Indeed, Jewish history cannot become raw material for abuse.

Meanwhile, X’s actions apply only in jurisdictions where “nudification” is illegal, while other platforms still remain poorly regulated. It means if users want to create and share sexualised images of women – Jewish or otherwise – they’ll find a way.

Time will tell whether the British government – and other governments – have truly grasped the depth of the problem.