Video · 8 min · Lesson 1 of 9
Where AI goes wrong
Five real kinds of incidents and what they have in common.
Transcript
Most AI incidents in companies are not dramatic. They are ordinary mistakes at scale: confidential data pasted into a free tool, an invented figure in a client report, a chatbot promising a refund the company never offered, biased screening of job applications, or an automated email sent to the wrong list.
What connects them is missing guardrails rather than bad technology. Nobody decided which tools were allowed, which data could be used, or which outputs needed a human check before leaving the company.
That is good news, because guardrails are cheap compared with incidents. The rest of this course covers the few controls that prevent most of these problems, and the legal framework that now makes some of them mandatory.
Key takeaways
- Most incidents are ordinary mistakes multiplied by automation.
- The root cause is usually missing guardrails, not the model.
- A few cheap controls prevent most problems.