On 28 May 2026, Appeals Centre Europe (ACE) — the Ireland-certified out-of-court dispute settlement body created under Article 21 of the EU Digital Services Act — published its first annual report. The headline number was striking: of the hate-speech disputes ACE reviewed between April 2025 and March 2026, it overturned the platform's decision to leave content up 70% of the time. TikTok fared worst, with 83% of reviewed hate-speech calls reversed, followed by Instagram (74%), Facebook (61%) and YouTube (58%). ACE received more than 24,000 disputes over the year — roughly one every 22 minutes — from users across the EU, including Dutch users for whom the Autoriteit Consument & Markt (ACM) serves as Digital Services Coordinator.
For a publication that favours proportionate regulation, the temptation is to dismiss the report as another EU compliance drumbeat. That would be a mistake. The strongest case for ACE is real, and worth stating plainly.
The case for the appeals body
Before the DSA, a European user who believed a platform had wrongly left up genuinely illegal content — a racist slur, an antisemitic incitement — had essentially no recourse short of national litigation. Article 21 created a free, fast, expert review channel. The examples ACE cites are not edge cases: racist comments comparing Black footballers to monkeys left up on Instagram after a Champions League match, and antisemitic videos circulated by prominent figures in Poland that YouTube declined to remove despite its own policy. When a platform's published rules forbid content and the platform leaves it up anyway, an independent second look is a reasonable, light-touch remedy. A 70% overturn rate is uncomfortable precisely because it suggests the platforms' own stated standards are being applied inconsistently.
That is the steelman, and it holds. The problem is what the rest of the report reveals about how those decisions were reached.
The default-decision problem
ACE can only overturn a platform if it can see the content. In a large share of cases, it never could. As ACE chief executive Thomas Hughes explained to Tech Policy Press, when a platform fails to supply the disputed material — because it cannot locate the content from the identifiers provided, or refuses to share it — ACE issues a default decision in the user's favour after 30 days. In the account-suspension category alone, ACE received more than 5,000 eligible disputes but issued fewer than 150 decisions after actually receiving the relevant content; the remainder defaulted (RTÉ, 28 May 2026).
This matters enormously for how the 70% figure should be read. A reversal reached after reviewing the post is an adjudication. A reversal reached because a platform's tooling could not retrieve a months-old comment is a procedural forfeit. Both land in the same statistic. The report is, in part, measuring platform data-management failures — and YouTube's year-long refusal to sign a data-sharing agreement, which Hughes called "disingenuous and misleading" — rather than pure moderation error.
Why the design should worry free-speech advocates
There is a deeper structural concern. ACE's decisions are non-binding — "an advisory decision," as DSA scholar Paddy Leerssen put it. Yet the report functions as a public league table that names and ranks platforms by how often they are "wrong." That creates a powerful one-way incentive: every default decision and every overturn points toward more removal, never less. There is no symmetric body cataloguing lawful speech that platforms wrongly took down at a regulator-adjacent body's urging.
Hate speech that is genuinely illegal under national law — the Netherlands criminalises group defamation and incitement under Articles 137c–137e of the Wetboek van Strafrecht — should come down, and proportionate enforcement of that line is legitimate. But "violates the platform's policy" is a far broader and vaguer category than "illegal," and the bulk of ACE's hate-speech caseload concerns the former. A system that issues 7,000-plus rulings against platforms, many without reviewing the underlying post, and publishes them as evidence of systemic under-removal, nudges platforms toward over-removal of lawful-but-awful speech to keep their numbers down. That is the classic collateral-censorship dynamic the DSA's drafters claimed they wanted to avoid.
A fixable system, not a failed one
None of this argues for scrapping Article 21. It argues for honest metrics. ACE should report adjudicated reversals separately from default decisions, so the public can distinguish moderation error from data-retrieval failure. Decisions reached without the content should carry a visible asterisk. And the inevitable enforcement attention now landing on platforms — the ACM in the Netherlands, Coimisiún na Meán in Ireland, and the European Commission for the largest platforms — should weigh under-removal and over-removal symmetrically, rather than treating a high overturn rate as unambiguous proof of platform failure.
The first annual report shows the appeals mechanism is being used, which is genuine progress for users. It also shows a process that, left as designed, measures the wrong thing well and the right thing poorly. Proportionate regulation means getting that distinction right before the league table hardens into a removal quota.