Aa
  • Read, Watch, Listen
  • TJI Series
  • Community Action
  • Events
  • Zedi
Follow us and continue the conversation
  • LinkedIn
  • Instagram
  • FaceBook
  • X
  • YouTube
Test yourself with the TJI Quiz!

Adjust size of text

Aa

Follow us and continue the conversation

  • LinkedIn
  • Instagram
  • FaceBook
  • X
  • YouTube

Your saved articles

You haven't saved any articles

What are you looking for?

Original reporting, insightful analysis and diverse views, delivered straight to your inbox

  • LinkedIn
  • Instagram
  • FaceBook
  • X
  • YouTube
Your support fuels our mission to inform, entertain and inspire.

The Jewish Independent is bound by the Standards of Practice of the Australian Press Council. If you believe the Standards may have been breached, you may approach The Jewish Independent or make a complaint to the Australian Press Council in writing at www.presscouncil.org.au. The Council may also be contacted on 1800 025 712.

  • Our Story
  • Contact us
  • Our Team
  • Support us
  • Guidelines
  • Pitch your idea
  • Republishing us
  • Community Action
  • Our Awards
  • LinkedIn
  • Instagram
  • FaceBook
  • X
  • YouTube
Aboriginal FlagTorres Strait Islanders Flag

The Jewish Independent acknowledges Aboriginal and Torres Strait Islander peoples as the Traditional Owners and Custodians of Country throughout Australia. We pay our respects to Elders past and present, and strive to honour their rich history of storytelling in our work and mission.

© The Jewish Independent 2026All Rights Reserved
Website design and development by Your Creative, MelbourneTerms & ConditionsPrivacy Policy
HomeRead, Watch, ListenAustralia
AnalysisAustralia

Content platforms grilled over online antisemitism

Executives from YouTube, Meta, and TikTok were quizzed about their efforts to remove antisemitic content, at the Royal Commission this week
Peter Kohn
Print this
Social media cauldron

Image: TJI

Published: 9 July 2026

Last updated: 9 July 2026

A glaring example of how antisemitic content can find its way past a welter of platform self-regulation and remain online came to light at this week’s sitting of the Royal Commission on Antisemitism and Social Cohesion.

The RC is now in its second week of Hearing Block 3, the media phase of the inquiry.

A video posted to YouTube after the Bondi Beach massacre remains online on YouTube today. In it, four males speculate that Arsen Ostrovsky, a senior official at the Australia/Israel & Jewish Affairs Council who sustained severe head injuries in the shooting attack, was “an intelligence asset”, “has a degree in theatre”, behaved “very crisis actor-ish” and used faked blood, and that it’s “false flag after false flag after false flag” and “Zionists do this kind of s**t all the time”.

That this video survives online placed a sobering boundary around claims by a top Australian YouTube executive that the platform’s removal rates of violative material have increased significantly in accordance with its own Community Guidelines.

It was a sit-up moment at the RC on Tuesday afternoon. Trying to explain YouTube’s decision on the Bondi video – and to counter the description of the video by Richard Lancaster SC, counsel assisting, as “a clear breach” of YouTube’s guidelines – Rachel Lord, the platform’s senior manager, government affairs & public policy, Australia/New Zealand, went on the defensive.

“What I have been told from the teams that have reviewed the video is that it is non-violative,” she insisted. She said YouTube’s policy is to remove videos that deny a violent attack took place or that people were harmed in it, but not videos that questioned motivations around it.

Lancaster countered: “I appreciate this isn’t your decision to make or unmake, Ms Lord, so don’t take this as a personal criticism, but I took you to the hate speech policy, which said in terms ‘we don’t allow content that promotes hatred on the basis of, among other things, being a victim of a major violent event’. So could it seriously be considered in your view that this does not violate that policy?” Lord then repeated YouTube’s regulations.

Grilling her on the fact that the video slurs Ostrovsky for being “Zionist” and “an intelligence asset”, Lancaster asked, “Surely that incites hatred towards [Ostrovsky] based on his characteristics as a Jewish Australian?”

Parsing the thinking of YouTube’s review teams, Lord observed the video didn’t contain material using “Zionist” as a coded word for a protected group (such as “Jews”), and that for YouTube,  “Zionist” in its original meaning is not classified as a protected group.

To that, Lancaster told her: “It may make matters worse, Ms Lord, rather than better”, adding, “You’re the representative who’s here on behalf of YouTube, Ms Lord. I suggest to you that it shows a really serious deficiency in YouTube’s guidelines that this video, after review at the highest levels, remains accessible to the public.”

The terse exchange graphically showed the fine-pencil method by which online platforms draw their regulations on curbing hate material, appearing to avoid catch-all scenarios that might conflict with other corporate priorities.

Meta’s tilt towards ‘free speech’

A “pivot” to favouring free speech over enforcement – and the unwieldiness of a worldwide user base of 3.5 billion – were issues that surfaced in close questioning of a senior Meta Platforms executive.

The change of policy on removing hateful posts was announced on January 7, 2025 (less than two weeks before the second-term inauguration of US President Donald Trump, a strident free-speech advocate). The switch has seen a decline in the number of hate posts being removed from Meta Platforms Inc’s social media outlets, Facebook, Instagram and Threads, the RC was told on Monday.

Testifying from the US, Benjamin Good, global director of Meta’s core policy team, explained Meta’s two main approaches to removing offensive posts – a proactive approach with a large AI component that removes material before it is published, and a reactive approach that relies on user reports to remove posts already published.

Good described proactive vetting as the “gold standard” but said it comes with “a significant risk of over-enforcement” and is used mainly on material instigating terrorism and violence.

Since the January 2025 “pivot” – a word used by Meta founder and CEO Mark Zuckerberg to describe that change ‒  Good said hateful content that doesn’t directly reference terrorism and violence is largely handled by the reactive approach. But some forms of antisemitism, such as those publicised by Hamas and Hezbollah, would be caught in the “proactive” net.

However, he accepted Commissioner Virginia Bell’s point to him that reactive vetting “can never achieve the gold standard” because those posts remain online until reported.

The RC heard Zuckerberg’s observation that the new policy would “catch less bad stuff” but that before the 2025 change, one-in-10 proactive removals, according to the Meta chief, represented “over-enforcement”.

Good conceded to counsel Richard Lancaster SC that “a plainly antisemitic post would be classified as less severe” and would be addressed only after a user report. But he said Meta’s policy is to allow content to remain on its platforms if it is similar to comments made in the US Congress or on mass-media news services.

Moreover, he said “over-enforcement” has been shown to sometimes remove posts actually opposing hateful posts – many from the Jewish community and a large number of those in the post-October 7 period.

Good was shown statistical evidence of a dramatic decline in removals of hateful posts in 2025 since the policy change (about 79 per cent, from 5.8 million items in the October-December 2024 quarter to 1.2 million in July-September 2025 and a 73 per cent decline on Instagram). He was asked by Lancaster whether he could give “a plausible reason” for the decline being due to anything other than the 2025 policy change. He responded that he had no great expertise in Meta’s measurement processes.

Questioned by Lancaster about the eSafety Commissioner’s proposal for a mandatory online hate code supported by civil penalties, Good would not commit to supporting such a code but would “welcome a discussion”. Asked more broadly if he accepts that an Australian regulator has the right to be able to enforce regulations on material viewed by Australians, Good said Meta respects Australia’s “sovereignty”.

Jewish community counsel Nicholas Bender SC put it to Good that reactive moderating means that Meta Platforms places the onus for content moderation on the victims of online hate speech themselves, such as the Jewish community. Furthermore, many Australian Jewish community members have reported to monitoring organisation, the Online Hate Prevention Institute, that they are suffering from reporting fatigue due to many reports not drawing any response from Meta.

Good said Jewish users’ frustration at not receiving responses from Meta is “terrible” for the Jewish community and “bad for Meta”, but that sheer scale made it challenging. At several points in his testimony, he referred to Meta Platforms’ 3.5 billion users, arguing “scale” is a continuous challenge to Meta in combating hateful content.

Good said statistics on the “prevalence” of hateful material indicate it is low, but that Meta doesn’t categorise hate material into specific types, such as antisemitism. This is despite a recommendation from its own Oversight Board. He said Meta “does not have a definition of antisemitism, and again, that’s really unfortunate because it is a very dynamic and shape-shifting problem”. (He described the IHRA definition as “a useful starting point”.)

The RC heard that an Instagram post making false and distorted claims about Auschwitz was allowed to remain up for three-and-a-half years. It had been posted in September 2020, one month before Meta explicitly prohibited Holocaust denial, and was reported six times and reviewed by two human reviewers, after which the Oversight Board intervened. Good acknowledged that the post should have been removed after the implementation of Meta’s prohibition of Holocaust denial posts.

The Jewish community’s weariness over unanswered reports perhaps explains why some try direct approaches to Australian arms of the platforms, where these exist. Facebook Australia’s director of public policy, Mia Garlick, testified that it receives  “a handful a week” of approaches to it outside the online reporting tool. It refers these to the escalations team at Meta Platforms in the US, but only hears back on some of these referrals.

Muffins, pizzas – coded antisemitism for kids

CyberWell, an independent Israel-based organisation dedicated to combating online antisemitism, briefed the RC on the main antisemitism categories it has devised – classic tropes about secretive control and power; Holocaust denial, distortion and mockery; conspiratorial self-victimisation (Jews are secretly responsible for their own misfortune); false-flagging (contriving attacks on Jews); and calls to violence against Jews.

Tal-Or Cohen Montemayor, CyberWell’s founding executive director, told counsel Nicholas Beshara SC that her organisation achieves “just above 50 per cent” removal on material it has escalated to platforms, but that’s because it is seen as a trusted partner. Individual user accounts reporting antisemitic material for removal can track as low as 21 to 9 per cent, she said.

She highlighted AI-generated “coded antisemitism” as an emerging trend. The RC was shown an Australian-designed cartoon on TikTok with the tagline, “It’s impossible to bake six million muffins.” She said words like cookies and pizzas are used to a similar effect. The veiled reference to Holocaust denial targets impressionable minors and manages to dodge the eye of AI and human content moderators, especially employees based in countries and societies with little understanding of Jewish history and culture.

Cohen Montemayor said the comments sections of social media and mainstream media platforms are “the wild, wild west” in terms of low levels of moderation, with Jews statistically more likely to be harassed there than in any other online space. For calls to violence after a violent attack on Jews, the least moderated comments section is that of Al Jazeera, she charged.

Platforms declining to categorise online hate comments by type means there are no comprehensive data sets on AI-generated antisemitic material, a necessary step, she said. “Only then are you going to be able to provide clear data, not only on the prevalence, but also how all this specific content is being recommended in our systems and what is the effective way of actually enforcing our policies at scale.”

Online hate ducks and weaves between bureaucrats

Finishing last week’s sessions on online hate, the RC heard about logjams between the federal government, the Department of Infrastructure and the eSafety Commissioner in the development of a regulatory framework for social media platforms – challenges that appear to be magnified by the absence of an Australian licensing scheme.

Earlier last week, the RC had heard testimony relating to extreme frustration from Jewish “lived experience” witnesses seeking to report antisemitic online activity.  But in her testimony on July 3, Sarah Vandenbroek, first assistant secretary, digital platforms, safety & classification division, at the Department of Infrastructure and Communications, reported that the eSafety Commissioner had expressed opposition to a “no wrong doors” approach.

Vandenbroek said the eSafety Commissioner also opposed a “warm handover” approach for streamlining the complaints process, which would involve the first contact briefing the referred contact, rather than passively referring the complainant. Vandenbroek said the eSafety Commissioner had expressed concerns that such cross-briefings would involve the handling of sensitive information across agencies.

Complaints to the eSafety Commissioner have doubled from 2025 to 2026; however, the 2023 Rickard statutory review into the Online Safety Act recommended not setting a numerical threshold on the volume of complaints. It preferred that the department develop an enforceable “digital duty of care” model.

After testifying that a volume threshold defining complaints would increase the volume to a level the department was not resourced to handle, Vandenbroek was pressed by RC Commissioner Bell on whether consideration had been given to increasing resources.

Vandenbroek responded that it was something that the government itself would need to recommend to the department. Bell then asked Vandenbroek how, in the absence of a licensing scheme for overseas social media platforms with an Australian presence, a digital duty of care could be applied to an uncooperative platform.

“I think it will be a case of designing the digital duty of care to be as effective as it can be,” said Vandenbroek. To which Bell responded, “As it can be ‒ for a company that has no physical presence in Australia and is not required to be licensed in order to operate here.”

“That is obviously a challenge, yes,” replied Vandenbroek.

Role of generative AI

Claude, the popular generative AI app based on a large-language model (LLM), has been trained to the technological equivalent of learning a language, explained Theo Hourmouzis, general manager of Anthropic PBC Australia, the Australian arm of the global Anthropic PBC corporation that developed Claude.

He said Claude is programmed to redirect harmful user requests with supportive messaging, and if a user persists, the chat is ended, and could even be escalated to the policy team for referral to law enforcement agencies. That regime operates in Australia, he added.

Under questioning by Lancaster, Hourmouzis assured the RC that pre-deployment testing of Claude’s 4.8 version had shown it could not be used to develop antisemitic language for use online. He said test prompting of a later real-world sample revealed that Claude, if pushed to its limits, could generate such material only in 0.01 per cent of the prompts tried on it, and Anthropic has sought to eliminate even that margin.

Although Hourmouzis was not aware of any Australian organisations combating antisemitism that Anthropic’s international corporation may have engaged with, he said it has partnered with the US-based Anti-Defamation League, using ADL-supplied expert information on antisemitism to refine the test prompts to Claude.

Educating platform users

In his testimony on Tuesday, Zachary Hecht, global head of policy, trust & safety, for TikTok, also referred to “scale” as a hurdle to the moderation of antisemitic content.

The challenge of tracking mega-volumes of content came up after Lancaster asked Hecht about a post that had been judged antisemitic by AI moderation but was subsequently allowed to appear after a human moderator intervened. (Only when an alert was received from Australian online watchdog Dor Foundation was the post examined by TikTok’s trust & safety team and removed).

Arguments of scale aside, TikTok’s policy of applying proactive moderation widely prior to publication appeared to be significantly more effective than the Meta Platforms policy of proactive removal only for violent/terrorist posts. Meta’s removal rate for material deemed antisemitic was 57.3 per cent, while TikTok’s rate was 88.8 per cent, the best of all the platforms reviewed by CyberWell.

However, under cross-examination by Jewish community counsel Nicholas Bender SC, the RC heard that TikTok doesn’t publicise statistics on the prevalence of violative content in ratio to the number of removals, or run sample sets, so it remains unclear what impact its rate of removals is having on its overall content.

And unlike Meta, TikTok does not have an independent oversight board, testified Valiant Richey, TikTok’s global head of partnerships.

On the upside, TikTok has globally taken a foot-forward approach. In Australia, it has partnered with the Sydney Jewish Museum and Dor Foundation for Incommon, a program that created online content promoting empathy and knowledge about Jews and Judaism, mainly to Gen Z.

“Trying to positively impact perceptions” is how Richey described it. (On its website, Dor states that in six weeks, Incommon reached 5.4 million young Australians, generating more than 175 million views and lifting perceptions of similarity between the user and Jewish people by 10 per cent).

READ MORE

ABC’s self-vetting like ‘marking own homework’ (TJI)

You might also like
  • Antisemitism, homophobia and a radical response to both
    Dennis Altman
  • Confronting Richard Flanagan and the myth of the Israel lobby
    Mary Garden
  • Court decision fails to understand when ‘Fuk Israel’ becomes antisemitic
    Jeremie Bracka
logomark.67a07ee3
We believe in the free flow of information

Republish our articles for free, online or in print, by using our Creative Commons licence

You might also like

    Woman with land and houses behind her
    NewsAustralia

    Australia to increase sanctions against West Bank settlements

    Deborah Stone 2
    funding handouts
    NewsAustralia

    Who will get the money from the Cranbourne windfall?

    Gideon Cohen
    Commissioner Virginia Bell
    Royal CommissionAnalysisAustralia

    Now is the time for the Royal Commission to deliver, not be politicised

    Dean Sherr 2

Comments

No comments on this article yet. Be the first to add your thoughts.

About the author

Peter Kohn

Peter Kohn

Peter Kohn is an Australian journalist who began his career at the Herald & Weekly Times. For 43 years, he was associated with The Australian Jewish News. For the past decade, he held the title of Senior Journalist at the newspaper. He is the author of two novels. Rachel’s Chance is the story of his family’s experiences as Jewish refugees in Shanghai during World War II, and View From A Sandcastle is based on his own story, growing up in Melbourne suburbia during the 1960s and 1970s.

More from the author

FeatureWorld

‘Your help is extremely needed’: Jewish aid group on the ground after Nepal flood

NewsAustralia

Strategies for cohesion final focus of Royal Commission hearings

NewsAustralia

Evidence of declining social cohesion and the efforts to turn it around

NewsAustralia

Hate preachers, polarised debate, counter-terrorism