About Platform Actions
A structured, source-verified public record of enforcement actions taken by platforms against independent publishers—specifically those where an AI system was reportedly involved.
Why this exists
Platform enforcement is opaque by design. Platforms do not explain which systems made a decision, how confident those systems were, or what evidence they acted on. When the enforcement is automated—carried out by an AI content classification system, an automated policy engine, or an algorithmic detection pipeline—the affected creator typically receives a generic notice and no meaningful appeal pathway.
The result is a systematic information gap: thousands of enforcement actions that affect creators' livelihoods and access to audiences, with no public record, no accountability mechanism, and no research infrastructure for documenting patterns.
Platform Actions fills that gap. It is a public-interest dataset modelled on Lumen (which covers DMCA takedowns) but scoped to AI-involved platform enforcement. The goal is to give journalists, researchers, and regulators a structured, citable record of what platforms are doing.
What makes it different
Berkeley Protocol alignment
Collection and verification practices follow the Berkeley Protocol on Digital Open Source Investigations —the same methodology used by human rights investigators and international law practitioners. Every record has an archive URL. Every inference is documented.
Structured data, not a blog
Records are schema-validated rows, not narrative posts. Each incident carries controlled-vocabulary fields: platform, action type, AI system involvement, verification level, confidence score, and source pairs. The schema is published and stable.
MCP query interface
Researchers and journalists can query the dataset in natural language via any MCP-compatible AI client. Ask: "Show me all Meta account disables from 2024 where AI detection was the reported cause." No SQL required.
Durable by design
Each meaningful batch of records is published as a dated GitHub Release artifact
(CalVer: 2026.06.12). Releases are milestone-triggered, not
time-triggered—a scheduled snapshot of an unchanged dataset adds noise without
value. Every source URL carries a corresponding archive URL so records survive
post deletion.
Launch case study: Meta
The seed dataset focuses on Meta because that's where the phenomenon is best documented — AI-generated content policies, automated visual similarity detection, and appeal processes that are opaque even by platform standards. The schema is platform-agnostic from day one: TikTok, YouTube, X, Reddit, and emerging platforms are all in scope.
Who runs it
Platform Actions is maintained by a single curator in the validation phase. The intake methodology, schema, and editorial policy are documented in the methodology page and the project's architecture decision records. The goal is to make the methodology transparent enough that any researcher could reproduce or audit the dataset without needing to ask.
No tracking
This site carries no analytics scripts, no third-party trackers, and no cookie consent banners. Traffic metrics are read from the Cloudflare dashboard. If you find any third-party request in the network panel that isn't a resource you loaded yourself, please confirm the request still appears with all browser extensions and plugins disabled before reporting it.
Contact
Factual corrections, source additions, and methodology questions: curator@platformactions.org
For erasure enquiries, see the data policy. The short answer: the right to erasure does not apply to this dataset under GDPR Art 17(3)(d).
Records currently visible are drafts under curator review. The database will be published when 50 records have been curated through the intake workflow and the schema is stable for 10 consecutive records. The methodology and data policy are final; the record count is not.