Companion videos

Connecting AI to your email, calendar, and CRM safely — companion videos

The moment an AI tool gets to read your inbox, untrusted text becomes executable. The article walks through OAuth scopes, dedicated service accounts, and human-in-the-loop checkpoints; these videos make the underlying attack model — prompt injection — concrete enough to take seriously.

Primary pick

10:57

What Is a Prompt Injection Attack?

IBM Technology

Jeff Crume's "buy an SUV for $1" example is the cleanest 10-minute explanation of why direct and indirect prompt injection are different problems, and why filtering can't fully solve either. It pairs directly with the article's argument that you need least-privilege scopes, a dedicated agent account, and a human in the loop on anything irreversible — not a cleverer system prompt.

What you should get from this: Distinguish direct and indirect prompt injection and why filtering alone is not enough.

Watch or know first: None — watchable cold.

AI Expert note: Treat this as conceptual guidance. Do not use real company data until permissions, retention, logging and human-review boundaries are clear.

Open video page

Also worth watching

25:11

OWASP's Top 10 Ways to Attack LLMs: AI Vulnerabilities Exposed

IBM Technology

Zooms out from prompt injection to the wider OWASP Top 10 for LLMs — insecure output handling, sensitive information disclosure, excessive agency — which is exactly the failure-mode catalogue you want in mind before you grant Gmail or HubSpot scopes to anything.

What you should get from this: Use the OWASP LLM risk categories to review tool access, output handling and sensitive-data exposure.

Watch or know first: The primary pick's injection framing helps; no security background needed.

AI Expert note: Treat this as conceptual guidance. Do not use real company data until permissions, retention, logging and human-review boundaries are clear.

Open video page