What happens when AI implementation goes wrong?
Corporate compliance officers always want their organizations’ speak up culture and processes to be as versatile as possible, ready to address any concern employees might raise. So, we’d be remiss if we didn’t consider how to apply that objective to the emerging risk of 2026.
The risk that your AI implementations go wrong.
We now have a steady stream of news headlines alleging exactly that: AI agents hacking into other companies without permission; AI “slop grenades” from employees who don’t evaluate AI-generated materials before passing them along to coworkers, slowing down work for everyone; employee uncertainty about who is accountable for which AI systems, because AI apps and agents are proliferating faster than IT teams can manage.
If your company’s AI ambitions veer off course, are your speak-up policies and processes ready to capture those concerns? Even more important, is your speak-up culture encouraging employees to raise such concerns at all?
The risks are real
First let’s consider the range of AI issues that might plausibly come across your hotline. They can be grouped into several categories.
First are complaints about AI policy violations that you (ideally) have on the books. For example:
- Employees using AI in ways that they shouldn’t, in violation of company policy or the law
- Employees not using AI in ways they should, or even sabotaging AI systems, because they fear for their jobs or otherwise object to using AI
- Employees using unauthorized AI, such via a personal laptop they set up next to their company-managed workstation
You might also receive reports about AI systems malfunctioning, which may or may not qualify as compliance concerns:
- AI systems misbehaving in specific ways: “Our new AI agents just hacked into our customers’ confidential data!”
- AI systems failing to meet performance expectations more generally: “The system keeps flagging legitimate transactions as non-standard items for human review.”
- AI consuming more resources than planned: “Have you seen how many tokens this system burns through? It’s going to blow our budget for the quarter by next week!”
At minimum, your hotline system should at least be ready to receive and triage these calls. For example, you’d want some sort of script that hotline call center employees (even if those “employees” are AI agents themselves) can follow to solicit useful information from the person submitting the report. You’d want a workflow to evaluate the nature of the issue and determine whether it should be routed to the IT team, HR, legal, or some other risk oversight function. (You could even use AI to help devise those scripts and workflows.)
What your hotline system shouldn’t do is immediately direct all such reports to the IT department. Many AI malfunctions or policy violations can have regulatory implications, so compliance teams need the ability to review these reports and then get involved as necessary.
Still, all of the above is just the mechanics of how to handle reports of AI trouble. There’s a deeper issue here that ethics and compliance leaders need to think about, too.
As usual, it’s about corporate culture.
‘We need to talk about the AI’
The deeper issue is that employees might be afraid to speak up about AI not meeting management’s expectations. After all, at many organizations, the push to embrace AI comes directly from the top. Who wants to risk their job by telling a gung-ho CEO that the AI doesn’t work, or the AI is causing problems, or that the CEO’s visions for AI transformation don’t square with how employees actually work?
That is a failure of culture and trust rather than tools and policies – and it can cause real trouble for companies trying to harness this powerful, fast-evolving technology.
Fear of speaking up is nothing new to ethics and compliance officers, of course. But on this issue especially – when companies are trying so many AI transformations, and the risks from AI “gone rogue” can escalate so quickly and severely – fostering a culture of trust is more important than ever before.
Compliance officers have an opportunity here to tell management, “Hold on; we need to be sure employees feel comfortable telling us when AI isn’t working, so we can correct our course before an AI failure takes us over a cliff.”
Moreover, compliance officers are ideally suited to help build that culture of trust. Consider these practical steps.
First, frame the risks in business language that management can understand. AI failures can bring cybersecurity meltdowns, costly civil lawsuits, operational disruptions, bad publicity, and regulatory investigations. When you frame misconduct risks as business threats rather than simple regulatory violations, management will be far more likely to listen to your input.
Second, focus on messages and training to speak up, perhaps even more than you focus on the hotline. You can have a fantastic hotline and case management program, but if employees don’t feel safe to speak up, they won’t use it. Conversely, if they do feel safe speaking up, they’ll do so through many channels – your hotline will be only one among many.
Third, work with IT and other risk management teams to close the policy and control gaps you do have. For example, employees might be afraid to raise concerns about AI because no one has ever clarified who is accountable for that AI agent’s behavior. Maybe your monitoring controls or business continuity processes aren’t robust. Weaknesses like that aren’t necessarily any one person’s fault, but by resolving them, you can help employees understand the rules of your AI road – and then they’ll feel more comfortable traveling it.
That is a failure of culture and trust rather than tools and policies – and it can cause real trouble for companies trying to harness this powerful, fast-evolving technology.
Conclusion
The good news is that many more boards and CEOs do now seem to understand that rapid AI adoption carries real risk.
For example, the CEO complaining about “slop grenades” several weeks ago originally made headlines in 2025 for pushing all employees to go full throttle on AI adoption. Even the AI tech giants themselves are telling workers to use AI efficiently rather than widely. Now that AI agents are also in the news for escaping their environments and causing real damage and regulatory violations, management is even more attuned to the threat.
The lynchpin to successful AI adoption will be employees telling management what does or does not work – and the lynchpin to that communication will be a culture of trust, so that employees are willing to raise those concerns.
AI with Boundaries: Balancing Innovation and Compliance
Your teams are already using AI. This webinar gives you a practical framework to harness its potential – while managing the risks.
