This past March, a copyright claim filed under the name of The Walt Disney Company, using a standard Gmail address, triggered a chain of automated enforcement actions that wiped out a Jewish creator’s monetization on Facebook. The appeal was denied in minutes.
No human reviewed it. The platform labeled her account a “repeat offender” based on a single post.
More than two months later, she cannot tag the CEO of the company that is destroying her account. She cannot tag The Lawfare Project, which defends Jewish creators in court. Both are active, public accounts. The restriction is specific to her page. She has since contacted both The Lawfare Project through their website and the Anti-Defamation League. She’s still waiting to hear back.
She’s not an isolated case. She’s a data point.
That she is me.
Facebook’s enforcement isn’t broken. For Jewish and pro-Israel creators, it is working exactly as the data shows it was designed to work. And while the platform is dismantling the accounts of people documenting antisemitism and Jewish history, the content calling Jews “demons” is generating millions of views.
What Meta decided in January 2025
The current situation has a specific start date.
On Jan. 7, 2025, Meta CEO Mark Zuckerberg announced Facebook, Instagram and Threads would end proactive content moderation and shift to a community reporting model, framing it as a correction to over-enforcement. He called it a “trade-off” between catching more harmful content and protecting innocent users from false removals. The ADL measured what that trade-off actually produced.
Before the change, antisemitic comments on the Facebook pages of Jewish members of Congress averaged 6.5 per day. After Feb. 4, 2025, they averaged 29.9 per day. Toxic comments overall increased by a factor of 13.
On some days, researchers counted 701 toxic comments across just 66 posts. In December 2025, Sens. Ben Ray Luján (D-N.M.), Martin Heinrich (D-N.M.) and Jacky Rosen (D-Nev.) sent a formal letter to Zuckerberg documenting the nearly fivefold increase in antisemitic behavior and noting that under the new policies, 277 million posts that had previously been correctly removed would now remain on the platform. The senators also flagged that antisemitic content staying on Facebook was being fed directly into Meta’s AI training models, potentially teaching those systems to reproduce and amplify what they were absorbing.
Meta’s own Oversight Board broke with the company in April 2025, publicly stating the rollback had gone too far. In March 2026, the board called the community-based replacement a “rickety” substitute that fails in crises and conflicts. The community notes that are supposed to catch what proactive moderation no longer does often take 15 or more hours to appear, long after a post has already reached the peak of its viral cycle.
What the algorithm is actually pushing
In March, the Antisemitism Research Center published a report called “Engineered Exposure.” Researchers used Instagram normally—the way any user would—and tracked content the platform’s own recommendation engine pushed directly to their feeds. Over 96 hours, they documented 100 antisemitic posts. Those posts collectively generated 5.3 million “Likes” and an estimated reach of 280 million users. Instagram didn’t surface this content because users searched for it. The algorithm delivered it.
Here’s what some of that content actually said.
One video, which received 191,000 “Likes” and 184,000 “Shares,” blamed Jews for creating child sex trafficking, orchestrating 9/11, assassinating JFK, engineering the bubonic plague, sinking the Titanic and masterminding the Atlantic slave trade. It called Jews demons.
A second video placed a Star of David on a demonic idol marked “666,” displayed Israeli and American flags on the idol, and filmed Iranian flags as the idol was destroyed and celebrated.
A third used Hebrew letters and Jewish symbols to claim Jews exercise hidden, sinister control over American institutions, framing Judaism itself as an inherently malevolent force.
Researchers also identified 12 AI-generated fake rabbi personas with a combined following of 2.1 million Instagram users. Each presented a distinct persona. All of them promoted antisemitic tropes about Jewish financial control. One account, “Rabbi Goldman,” had 1.4 million followers with some videos reaching 5 million views. These accounts have since been traced to scammers in South India using fabricated Jewish religious authority to build monetized audiences. The hate content was a marketing funnel.
In April, the ADL released its report “How Meta’s Content Moderation Changes Risk Turning Instagram into a Hub for Hate.” Researchers systematically reported 253 pieces of violative content through Instagram’s standard reporting system, including accounts and posts linked to white supremacist networks, designated foreign terrorist organizations, and vendors selling Nazi merchandise. Instagram removed 11 accounts and eight posts. In 20 cases, Instagram explicitly stated it lacked the bandwidth to review the reports. The platform’s removal rate: 7%.
“When a platform used by 80% of American adults under 30 allows pro-Hitler content to rack up millions of views, Holocaust denial to spread unchecked and terrorist organizations to fundraise openly, we’re not talking about a policy disagreement,” said Oren Segal, senior vice president of the ADL. “We’re talking about a public-safety crisis.”
The copyright weapon
While antisemitic content accumulates millions of views, a separate enforcement mechanism has been documented targeting creators who cover Jewish history and counter-extremism. It runs through Facebook’s copyright system, and it’s being abused.
The DMCA’s notice-and-takedown framework requires platforms to act first and investigate later. A claim is submitted. Content is removed. After that, the creator hits a wall. There’s no visible step requiring verification of the claimant. No confirmation of authority. No escalation if the claimant never responds. Anyone can file.
Meta’s own rights management tools have been documented as vectors for bad actors: Bloomberg reported on scammers using the system to file takedowns over content they didn’t own, with some admitting directly to creators that no actual copyright infringement existed before demanding payment to stop future fraudulent claims.
As it pertains to my case, I received a copyright violation notice listing The Walt Disney Company as the rights holder. The contact email provided was a standard Gmail address. Disney’s legal team doesn’t use Gmail. Facebook’s automated system accepted the claim anyway, demonetized my page, and when I appealed on my birthday, denied it in minutes. No human reviewed it.
The same AI that processed my appeal then informed me that my page had been designated a “repeat offender” for community standards violations, based on that single post, with no prior violations cited.
I have rebuilt my following to nearly 50,000 after watching it get stripped. Since May 20, my professional dashboard has been toggled off repeatedly without me doing it. My subscriber feature has been shut off completely with no notice. My analytics are frozen, locked to a single week regardless of what date range I select.
I am a Meta Verified customer, a subscription sold on the promise of premium support and human review. I spent two hours and 15 minutes in its support chat recently and cycled through three agents, all of whom either transferred me or left the conversation without resolving anything. The last one promised a specialist team transfer and then disconnected.
The structure that enables it
Meta’s creator support doesn’t run on Meta employees. The company outsources its moderation and creator support operations to contractors through firms including Majorel and Teleperformance, which operate moderation hubs across Africa, Asia and Eastern Europe.
Investigative reporting from the Bureau of Investigative Journalism and Foxglove documented conditions at these facilities, including surveillance, coercion and workers being threatened with job loss for raising concerns. The agents staffing creator support chats follow scripts. They don’t have the backend access or the authority to fix account-level problems. When a case exceeds their script, they transfer. When the transfer fails, they leave.
Zuckerberg controls this system from a position that’s structurally insulated from accountability. Through a dual-class share structure, he holds 61.2% of Meta’s voting power despite owning roughly 13% of its equity.
In May 2025, a shareholder proposal demanding a report on how Meta combats antisemitism received support from 46.8% of independent shareholders, the highest level of support for any human rights-related shareholder proposal at any U.S. company that year. Both major proxy advisory firms endorsed it. Zuckerberg voted it down.
Meta reported $200.97 billion in revenue in 2025 and $60.46 billion in net income. It’s projected to spend between $115 and $135 billion on AI infrastructure in 2026. The company has been hollowing out human moderation to fund that investment, replacing it with automated systems that the April ADL study found were removing 7% of reported extremist content.
The asymmetry
Facebook’s AI is sensitive enough to identify which specific user will buy a particular product based on a $2 ad targeting signal. The same infrastructure that powers that precision is the infrastructure that handles content enforcement. The argument that Meta can’t find Holocaust denial or blood libel with the tools it’s built isn’t a technical claim. It’s a policy choice.
Meta’s enforcement responds instantly when the institutional exposure is high enough. The company removed over 90,000 posts to comply with government takedown requests, in an average of 30 seconds per post, according to internal data reviewed by Drop Site News in April 2025. Thirty seconds, when they want to move.
Yet for a Jewish creator whose subscriber feature has been shut off, whose analytics are locked and whose professional dashboard keeps getting disabled, Meta sends contractors who leave mid-conversation.
The legal horizon
The platform’s period of unchallenged immunity is ending.
In March, a New Mexico jury found Meta liable for endangering children and ordered $375 million in civil penalties. A Los Angeles jury, the following day, found Meta 70% responsible for harm caused to a minor addicted to social media. In August 2025, families of victims of the Hamas-led terrorist attacks in southern Israel on Oct. 7, 2023, filed a class-action lawsuit in Tel Aviv seeking approximately $1.1 billion, accusing Facebook and Instagram of enabling the Hamas massacre in real time by allowing livestreamed footage of murders and abductions to circulate on their platforms.
Legal analysts note that if the per-user damage framework from the New Mexico verdict were applied across pending cases in Florida and New York alone, Meta’s exposure could reach $40 billion. If a platform can be held liable for what its algorithms do to children in the context of sexual exploitation, the legal framework for algorithmic liability on antisemitism is closer than it looks.
What has to change
The ADL’s 2023 platform study established that YouTube, using the same algorithmic infrastructure as Facebook and Instagram, didn’t reward test personas who engaged with conspiracy-adjacent content by recommending increasingly extreme material. The other platforms did. YouTube proved the capability exists. The decision not to deploy it is a choice.
Structural reform means recommendation systems that categorically exclude antisemitic content from amplification before it goes viral. It means mandatory human review for account-level enforcement decisions that affect a creator’s livelihood. It means public removal rates, platform by platform, content category by content category, with independent verification. It means revisiting Section 230’s near-blanket protection for platforms that make deliberate algorithmic choices to amplify hate for engagement.
And it means advertisers whose brands are running adjacent to videos calling Jews demons need to know that’s where their money is going.
Meta generated $201 billion last year. It chose what to do with that infrastructure. Jewish creators documenting history are being dismantled by automated systems while content calling for Jewish extermination sits at millions of views. That’s not a malfunction. That’s what the company decided.
