A body set up to make Europe’s content censorship regime work has accidentally documented why it doesn’t.
Appeals Centre Europe, an Ireland-certified dispute settlement outfit operating under the EU’s Digital Services Act, released its second transparency report this week.
The numbers it published describe a system failing in both directions at once, and they hand the case against laws like the DSA to anyone who wants it.
Let’s start with what the body found when it actually got to look at the disputed content. Across the year from April 2025 to March 2026, it disagreed with the platform’s call 59 percent of the time.
Reclaim Your Digital Freedom.
Get unfiltered coverage of surveillance, censorship, and the technology threatening your civil liberties.
Break that down and the picture gets stranger. When users challenged content that platforms had deleted, the Appeals Centre sided with the user 52 percent of the time.
When users flagged content that the platforms had chosen to leave online, the body overturned that decision 63 percent of the time. The same companies are deleting things they shouldn’t and keeping up things the regime says they should remove, often in the same reporting period.
The machinery the DSA built to produce correct moderation outcomes is producing roughly a coin flip. Legitimate posts get censored. The body reviewing the censorship then has to tell the platform to put them back. More than half the time, when it can see the evidence, it concludes the platform got it wrong.
The Appeals Centre received more than 24,000 disputes over the year, with eligible cases arriving nine times faster in March 2026 than in April 2025.
That is the scale of disagreement a single dispute body is fielding from across the EU. It is also a fraction of the moderation decisions these platforms make every day, which run to millions.
The DSA’s underlying premise is that platforms can review this firehose of human expression and arrive at defensible, appealable judgments about each piece. The error rate on the small sample anyone actually checks suggests the premise was never sound.
Then there is the question of whether any of it gets enforced and here the report stops being merely damning.
Account suspensions are where the system collapses outright. The Appeals Centre received more than 14,000 suspension disputes.
It managed to fully review fewer than 150 of them, because platforms would not hand over the content needed to assess the bans.
More than 7,300 disputes ended in what the body calls “default decisions,” meaning the platform failed to supply the material within 30 days and the ruling went to the user automatically. Meta was the standout.
Out of more than 4,600 eligible Facebook and Instagram suspension disputes, the company produced the disputed content in fewer than 100 cases.
So a European can be banned, file a free appeal under the law that was sold as their protection, win that appeal by default because the platform never engaged, and still find their account gone.
The Appeals Centre says that even after it reviews cases and rules, platforms often don’t act on the outcome. In one section covering disputes from civil society groups, the body says it knows of only “a handful” of cases where platforms acted on its rulings, with many disputed posts staying online.
An appeals process that produces rulings no one is obliged to honor is nothing more than a complaint box with a logo.
Thomas Hughes, the body’s CEO, positioned the findings around enforcement failures. “Online hate and harassment have real-world consequences for many people and communities. In more than two-thirds of our decisions about hate speech, we found that platforms failed to enforce their own policies and left up hateful content. This goes to show that platforms don’t always get it right. If you’re in the EU, you can challenge a platform’s decisions free of charge to Appeals Centre Europe and get an expert, impartial review,” he said.
Read the same data without the institutional framing and a different lesson appears. “Failed to enforce their own policies” describes platforms not deleting content that someone, somewhere, has decided is “hate speech.”
The hate speech category the report leans on is exactly the kind of definition that depends entirely on who holds the pen. The body overturned platform decisions to leave up reported hate speech 70 percent of the time, with TikTok at 83 percent, Instagram at 74, Facebook at 61, YouTube at 58.
Those numbers can be read two ways. One reading is that platforms are derelict. The other is that a vast share of what one reviewer calls hate speech, the platform’s own systems looked at and decided was permissible expression. The disagreement is proof that nobody involved can agree on where the line sits, which is what happens when the rule is a category rather than a law.
Brussels spent years promising users a robust system for contesting what happens to their speech online. The body running that system just published the evidence that it doesn’t work and the deepest problem isn’t administrative. You cannot build a fair, reviewable, rights-respecting censorship regime on top of a volume of speech this large. The mess is the message.
