2023 · Prompt Injection
Chevrolet Chatbot Sells Car for $1
Cost: Estimated $80,000+ in brand damage and PR response
Read the case and prevention steps67 checks to catch hallucinations, prompt injection, data leaks and unsafe tool use before they reach your customers.
Open the checklistFree · printable · no signup
10 categories, aligned with the OWASP Top 10 for LLM Applications and for Agentic AI. Every item has a severity.
Catch when AI makes up facts, cites non-existent sources, or confidently delivers wrong answers
Test if users can manipulate AI to bypass rules, leak data, or behave maliciously
Critical security tests for data leakage, PII handling, and API vulnerabilities
Ensure AI can’t be tricked into generating harmful, offensive, or brand-damaging content
Catch formatting errors, broken logic, missing citations, and inconsistent responses
Test for demographic bias, stereotype reinforcement, and unfair treatment patterns
Safeguards against illegal content, brand violations, and regulated advice
Ongoing checks to catch failures in real-time before they go viral
Critical checks for AI agents that call tools, use MCP, or take autonomous actions (OWASP Agentic AI Top 10)
Scan AI-generated UIs, copy, and e-commerce flows for deceptive design patterns that manipulate users
Four of the 70 documented incidents in our database. Each one maps to a category above.
2023 · Prompt Injection
Cost: Estimated $80,000+ in brand damage and PR response
Read the case and prevention steps2024 · Hallucination
Cost: $812 CAD in damages plus legal fees
Read the case and prevention steps2024 · Jailbreak
Cost: Millions in negative PR, temporary loss of chatbot service
Read the case and prevention steps2023 · Hallucination
Cost: $100 billion in market cap loss
Read the case and prevention stepsRun all 67 checks before launch. Treat any failed critical item as a release blocker.
Re-run the checklist when you change the model, the prompt or the data your agent retrieves. Small edits can undo earlier fixes.
Spot-check production regularly. Models drift and attack techniques change.
Not every item applies to every agent. A support chatbot and an agent that calls tools need different emphasis; start with the critical items that apply.
Yes. No credit card, no email gate. Open it, print it, share it.
No. Each item has a plain-English explanation and an example test. Some items, such as tool-use checks, are easier with an engineer.
It depends on your agent. A simple FAQ chatbot is much quicker to check than an agent that calls tools or takes actions. Start with the critical items.
Yes, please do.
A clean, printable web page you can save as PDF from your browser.
No checklist can. It covers the failure modes behind well-documented incidents, so you can find and fix the common ones before your customers do.
Open the checklist, print it, and work through the critical items before your next release.
Open the checklist