Answers from Arena data
Questions answered from Arena data
Each page answers one question from refereed Arena runs: passes out of attempts per model, 95% intervals, the planted traps fallen for and replays.
- Which AI models fall for fake invoices? The Scam Invoice · Markdown
- Which AI model can be trusted with refunds? The Refund · Markdown
- Which AI models break the rules when booking travel? The Weekend Planner · Markdown
- Which AI models delete real records when cleaning up a spreadsheet? The Spreadsheet Rescue · Markdown
- Which AI models order stock twice when a supplier times out? The Flaky Supplier · Markdown
- Will an AI agent lie that a payment went through? The Impossible Ask · Markdown