Don’t Just Give AI a Task – Give It Boundaries
The more responsibility you give AI, the more important it becomes to tell it what it must not do.
Last week’s AI security incident provided an extreme example: autonomous AI agents pursuing an objective went beyond the boundaries their developers expected.
The same principle applies to everyday business use.
When giving AI an important task, don’t simply specify the objective. Add guardrails defining the limits within which it should operate.
Instead of:
“Review these customer accounts and recommend which ones should have their credit limits increased.”
Try:
“Review these customer accounts and identify potential candidates for a credit-limit increase. Do not make or implement any credit decision. Do not use protected or irrelevant personal characteristics. Flag incomplete or conflicting information. Explain the evidence behind each recommendation. Any final decision must be made by an authorised employee.”
The second prompt doesn’t just tell AI what to do.
It establishes:
The task → The boundaries → The exceptions → The human decision point
Why this matters
As AI moves from answering questions to performing increasingly complex business tasks, good prompting isn’t only about getting a better answer.
It’s also about establishing control.
A useful habit is to give AI four things:
1. Objective — What do I want you to achieve?
2. Boundaries — What must you not do?
3. Escalation — When should you stop and ask me?
4. Authority — What decisions remain with a human?
Executive takeaway
The more autonomy you give AI, the clearer its guardrails should become.
You wouldn’t give a new employee unlimited authority simply because they were capable of doing the job.Don’t give it to AI either