Insights

Governance

Four things your board calls one question.

"Is our AI secure?" is really four questions. Your board will ask the first. Ask yourself the other three.

Rujuta Singh

Rujuta Singh

3 min read

Four different things can go wrong with your AI. Your board calls all of them one thing: "Is our AI secure?"

One. Can it be attacked?

Someone hides instructions in a web page or a PDF. Your AI reads it and follows them. That's prompt injection. OWASP ranks it the top risk for AI applications.

Two. Can you trust it to act?

A wrong answer is annoying. A wrong payment is a meeting with your CFO.

Three. Can you prove what it did?

"The AI did it" won't satisfy an auditor. You need the record, and proof a human approved what mattered. Plenty of teams have the AI. The audit trail is still on the to-do list.

Four. Are you allowed to run it?

The EU moved its high-risk AI deadlines to December 2027 and August 2028. The transparency rules already started in August 2026. More time still has a deadline at the end of it.

Your board will ask question one. Ask yourself the other three.

None of this is a reason to stop building. It's a reason to build with your eyes open.