An AI agent is an AI that can take actions, not just answer questions. One with access to your live systems can change real things at machine speed.
Overreach is not always an attack. Often it is an honest misunderstanding: asked to fix a small bug, the agent decides the right fix is to rebuild part of the database and publish it.
What works: narrow access, separate keys for each job, no direct access to live systems, a person approving anything that deletes, and a record you can read afterwards.
Make pausing and cancelling an agent as easy as starting one.
Words you will see
- Agent
- An AI that can take actions, not just answer questions.
- Audit trail
- A record of what was done, by whom and when.
- Approval gate
- A required human "yes" before a risky step.
Decision
An AI agent changed your database and put it live. A customer reports missing data. What first?
What to do instead
- Give each agent only the keys for its own task.
- Have a person approve anything live or anything that deletes.
- Keep a readable record of every action.
Common mistakes
- Letting an agent connect to the live database.
- Believing a written instruction is a safety control.
- Taking the agent's own summary as the truth.
References and further reading
- Fortune: Replit agent database deletionFortune
- OWASP Top 10 for LLM applicationsOWASP GenAI Security Project
Finished this one?
Add it to What I learned and earn 50 Trust. You can open your list from the badge at the bottom of the page.