Guardrails
Also known as: AI guardrails
What is Guardrails?
Technical controls that keep an AI system's behaviour inside agreed limits, checking inputs and outputs and blocking anything unsafe before it reaches a user: a leaked secret, a harmful reply, an off-topic tangent. Dedicated frameworks let teams define these rules in code. Guardrails enforce policy at runtime. An AI policy writes it down.
Why it matters
Guardrails are easy to add and hard to get right, and the difficulty is a balancing act nobody escapes. Set them loose and unsafe output slips through, which is the whole thing you were guarding against. Tighten them and they start blocking legitimate use, so people hit dead ends on reasonable requests and write the tool off as broken. Neither failure shows up until someone measures it. There is a maintenance cost as well, easy to forget. The situations the model faces keep shifting, new ways to misuse it appear, and a guardrail written once and left alone slowly stops matching reality.
In practice
A customer-facing assistant refuses an ordinary question because a broad filter caught an innocent word, and the customer leaves with nothing. Meanwhile a carefully phrased harmful request gets through the same filter. The real work is continuous tuning against live traffic: watching what the guardrail blocks, what it lets past, and adjusting both, rather than shipping one ruleset and assuming it holds.
