3.7 Million Scam Accounts and the Cost of Siloed Systems

Meta, together with the Singapore Police Force, stopped 3.7 million scam accounts, pages, and content items.

Formally, this is a report on combating scams.

At its core, this demonstrates that one company's moderation cannot catch organized crime: the network operates across dozens of platforms and jurisdictions at once, and without data exchange between systems, the full picture remains invisible.

For business, the lesson goes beyond anti-fraud: the gaps between siloed systems where scammers hide on Meta consume budgets inside ordinary companies through duplicate suppliers, fake leads, and duplicate payments. The schemes blocked by Meta and SPF are familiar and standard: investment chats promising guaranteed returns with zero risk, premium cosmetics stores offering discounts on pages created a week ago, and trading signal groups.

Every scheme starts with one message and one link, but relies on infrastructure consisting of dozens of linked accounts and payment chains.

One detail from the report stands out: some networks attacked users in several countries at once through localized versions of the same cover story—they changed the language and brand name but kept the scheme unchanged.

This is addressed with an engineering layer around the data.

An MDM/PIM platform (Pimcore, Akeneo, Riversand) maintains the canonical description of an entity—an item, supplier, or customer: each record has one source of truth instead of five versions across five systems.

An integration bus (MuleSoft, Talend ESB, Apache Kafka) transmits signals between systems in real time.

A nightly batch will not catch such a network: while the signal waits for the next cycle, the fraud has already shipped. LLM & Security Gateway gives AI agents access to sensitive data for pattern matching without exposing the entire database—the same principle by which Meta shares indicators with SPF rather than raw user data. RAG and MCP let the model match signals across different repositories without physically moving data into one place: the search runs through an index while the repositories stay where they are.