Extend Verify / document fraud detection API

Fraud Arcade

Extend Verify detects documents with altered content, tampered with AI down to a single character, or generated from scratch. Five documents: tell which are legitimate, then see the evidence.

  • 5 rounds
  • Legitimate or fraud
  • About 3 minutes

  1. ( 01 / 03 )

    Inspect

    Read the page, the story behind it and the file metadata.

  2. ( 02 / 03 )

    Call it

    Legitimate or fraud. Say what gave it away if you can tell.

  3. ( 03 / 03 )

    See the evidence

    The reveal circles what Verify flagged and what it would have cost.

About this game

AI makes a convincing forgery trivial to produce and nearly impossible to catch by eye. Fraud Arcade is a 5-round game built on Extend Verify, the document fraud detection API built for developers and agents. Each game deals 5 documents from a set of 20 real invoices, receipts, statements, forms and letters. Each one is either the untouched original or a tampered copy prepared by the Verify team.

Half of the tampered copies are metadata-only fraud, such as a value rewritten inside the PDF or a revision saved after the document was created; Verify Light caught these. The other half are content changes: one value edited so the document contradicts itself, such as a total that no longer adds up or dates out of order; Verify Agentic caught these.

Read the context, inspect the page and its metadata, then call it legitimate or fraud. The reveal circles the changed value, shows how Verify caught it and which Verify mode caught it, and explains what the fraud would have cost downstream.

Play through your agent

Agents play the same rounds through the Fraud Arcade MCP server (Streamable HTTP, no auth) at https://arcade.extend.ai/mcp. It has seven tools: instructions, play1 to play5 (one document per round, with a play_ID and file_ID) and results (takes the agent's five calls and returns the answers with Verify's explanations). No answer is revealed until all five calls are in, and nothing is stored.

  • Claude Code: claude mcp add --transport http --scope user fraud-arcade "https://arcade.extend.ai/mcp"
  • Codex: codex mcp add fraud-arcade --url "https://arcade.extend.ai/mcp"
  • Cursor: add to Cursor

Then ask the agent: “Play Fraud Arcade using the fraud-arcade MCP server. Call its instructions tool first and follow it exactly.” For the Extend MCP server that parses, splits and extracts your own documents, see the MCP server documentation.

Extend Verify

Verify detects fully AI-generated documents, AI edits as small as a single character, changes made after signing, and patched text, in-painting and overlays made with Photoshop or PDF editors. It looks for signs of fraud within the document itself; it does not verify identity or assess external risk signals. It returns detected signals, a low, medium or high risk assessment, explanations and traceable evidence. Metadata-only fraud is caught by Verify Light; content changes are caught by Verify Agentic. Verify supplies the evidence; your workflow determines the action.

Verify is available today, upon approval. Request access or read the launch post.