AI Automation Without Guardrails Is Just Hope
- gwen sparks
- Feb 9
- 3 min read
Why every system needs someone minding the store
The problem didn’t announce itself as a crisis.
It showed up as a helpful answer.
In 2024, an Air Canada customer asked the airline’s chatbot about bereavement fares. The chatbot responded confidently, offering guidance that turned out to be wrong. The customer relied on it. When the airline later denied the fare, Air Canada argued that the chatbot wasn’t really the company, just a tool providing information.
The court didn’t see it that way. The ruling was blunt: the airline was responsible for what its AI said. The chatbot wasn’t a side experiment. It was speaking on the company’s behalf. (American Bar Association summary)
Nothing about this failure was technical. The system worked as designed. What failed was ownership. No one had clearly decided who was accountable when the AI spoke with authority and got it wrong. And once that answer went public, the explanation arrived too late. That’s the quiet risk most organizations are still underestimating.
This wasn’t a one-off
Air Canada isn’t an anomaly. It’s one of the cleaner examples.
Similar moments have surfaced across sectors, from customer service chatbots confidently inventing policies, to automated decision systems issuing harmful outcomes at scale, to facial recognition tools producing biased results that organizations struggled to explain after the fact.
Different industries. Different tools. Same pattern. AI doesn’t need to malfunction to cause damage. It only needs to operate confidently while accountability stays implicit.
Why communicators feel this first
When AI creates confusion, it rarely shows up as a technical failure. It shows up as a message problem. A customer wants an explanation. A journalist asks who approved something. A leader needs to stand behind a decision they didn’t personally make. That’s when communicators get pulled in.
We’re asked to explain:
why an AI said something it shouldn’t have
why a decision happened without a clear owner
why a system was trusted without visible oversight
By the time communicators get involved, the output is already public and the stakes are already human.
This is why AI governance conversations often feel abstract until they aren’t. Communicators don’t encounter AI as a tool. We encounter it as a voice, a decision, or a moment that needs explaining. And when ownership isn’t clear, explanation becomes improvisation.
This isn’t a tools problem
Most organizations still frame AI as a capability question:
Which tools do we allow? Which models perform best? Who gets access?
Those questions matter. They’re also insufficient. Because tools don’t damage trust. Unowned decisions do.
You can’t rely on people to “use good judgment” at scale. Judgment without structure collapses under pressure, especially when systems move faster than human review. At some point, most organizations reach the same realization:
We trained people. We set guidelines. Why does this still feel fragile? That’s why AI needs supervision, not just implementation.
The missing layer: accountable governance
Governance often gets dismissed as bureaucracy. But real governance isn’t about control. It’s about clarity. Clarity on:
who owns AI systems end-to-end
who can deploy, pause, or override them
what risk is acceptable and who decided that
how decisions are documented and reviewed
Without that clarity, organizations aren’t managing AI. They’re hoping it behaves.
Where ISO/IEC 42001 fits
ISO/IEC 42001 is a global standard that serves as a playbook for how organizations should govern AI, not at the tool level, but at the decision and accountability level. Organizations can choose to align with the standard, and some may pursue formal certification. It exists for a simple reason: intention doesn’t scale.
Most organizations don’t struggle because they lack principles. They struggle because, under pressure, no one knows who decides.
ISO/IEC 42001 isn’t about picking models or approving tools. It’s about answering the questions that surface after something goes wrong:
Who was responsible for this system?
Who had the authority to deploy it?
Who could have paused it?
Where was that decision documented?
What was supposed to happen when it failed?
In practice, the standard pushes organizations to make those answers explicit before they’re needed. It encourages teams to agree, in advance, on ownership, oversight, and escalation so responsibility doesn’t disappear when systems move fast or teams change. Not because AI is inherently dangerous. But because confidence without supervision creates risk faster than most organizations can respond. (ISO/IEC 42001)
The quiet truth
If your AI strategy depends on people being careful, it isn’t a strategy.
Automation without supervision is just hope. Tools make AI possible. Clear ownership makes it trustworthy. And the moment AI starts speaking for your organization is the moment someone needs to be unmistakably minding the store.
Comments