How confident do you feel about your organization’s AI implementation? This question matters because nearly half of employees admit to using “shadow AI” to meet their work obligations. While employers may be worried about security, they shouldn’t ignore unsanctioned AI’s impact on performance and trust.
Research from the cybersecurity vendor BlackFog found that 49% of employees reported using AI tools at work that hadn’t been sanctioned by their employers. So, why is this happening and what are the implications?
What is shadow AI?
Don’t worry, shadow AI isn’t as terrifying as it sounds. It refers to any AI tool that an employer hasn’t officially implemented or otherwise sanctioned for use in their business. In most discussions around shadow AI, cybersecurity seems to be the main concern.
This is pretty understandable. Unregulated AI won’t necessarily have the same safety nets or level of encryption that fully tested, professionally implemented tools can offer. But that’s not what we’re here to talk about.
The real problem with unsanctioned AI tools
While the potential for shadow AI to disrupt your cybersecurity is important, it’s far from the only reason to address unsanctioned use.
It may not be the first thing you consider, but shadow AI also poses a major risk to your workplace’s operating model. Left unchecked, it can damage trust and impair your ability to assess employee performance. So, why is this the case?
How shadow AI affects visibility in your organization
When you use tools (AI or otherwise) that sit outside of established workflows, it diverts information away from the systems that need it. This can include:
- Performance notes that contextualize project progression and employee development.
- Feedback for personal growth that should sit inside performance management.
- Recognition that would otherwise give colleagues visibility for their work.
When all this information sits outside of official systems, it becomes fragmented and easy to lose track of. When it’s time for a performance review or project debriefing, you get an incomplete picture that’s missing essential insights.
How this affects performance, trust, and retention
Addressing the problems of shadow AI is unfortunately more complicated than simply gathering the relevant information from unregulated tools. Even if it wasn’t, it would be a terribly inefficient use of a manager’s time. Here are some of the issues HR must contend with:
- Low quality performance conversations: When managers can’t surface complete employee performance records, they’re left to muddle their way through reviews. Meanwhile, incorrect feedback summaries and incomplete goal tracking give employees the wrong idea of where they stand.
- Compromised recognition processes: Missing performance notes and fragmented recognition lead to employee efforts becoming invisible. This will disproportionately impact your quietest contributors and hybrid workers who aren’t a regular part of office culture.
- Accelerated disengagement and turnover: When your people feel their efforts aren’t being acknowledged, it’ll inevitably affect their dedication. Employees will lose faith in recognition processes and stop taking performance conversations seriously. As disengagement rises, employee retention becomes much more difficult.
These issues make it much more difficult for HR to maintain reliable metrics for engagement and performance. In the short term, this means suboptimal work and mounting disengagement. Long term, it threatens HR’s ability to predict trends, inform stakeholders, retain talent, and support employee productivity.
Why your people turn to shadow AI
The question is, if shadow AI causes so many problems for both its users and HR, why have so many employees embraced it?
It’s important to remember that employees using shadow AI aren’t doing so to cause problems. Quite often, they’re using it for the same reasons people use any AI tool: cognitive offloading, streamlining tasks, and increasing personal productivity.
The survey from BlackFog found that 69% of C-suite leaders believe speed and efficiency trump security concerns. In that sense, the willingness of employees to use unapproved tools is a response to the pressures of increasing work demands.
BlackFog’s survey also found that 63% believe it’s acceptable to use unsanctioned AI tools without IT oversight if their employer hasn’t provided a suitable alternative. Furthermore, 21% believe employers would look the other way as long as deadlines were met. So, it seems like employees are more likely to use shadow AI in response to ambiguous policies and low levels of enforcement.
Finally, employees may feel pressured to conceal AI usage (shadow or otherwise) for fear it would devalue their contributions. Not only does this imply a lack of trust in employers, it can also damage managerial trust at work by making it unclear how employees achieve results.
How to bring work out of the shadows
It’s important to remember that this is an issue of visibility above all else. Nobody’s asking you to prevent AI use. Instead, HR must design systems and solutions your people don’t feel the need to work around. Let’s look at how you can pull this off:
Treat shadow AI as a signal and normalize AI use conversations
Besides security concerns, shadow AI is rarely a compliance problem. It’s usually a signal that work is evolving faster than your systems. Instead of treating AI use as something to police, normalize open conversations about where and how people are using it. Regular check-ins, team retrospectives, and manager 1:1s create safe spaces to surface AI-assisted work without fear.
Visibility improves when people feel supported. When they feel comfortable bringing up the gaps in approved AI tools or suggest new ones, it eliminates a lot of the pressure to rely on shadow AI.
Create real-time performance visibility
When performance is only visible at review time, people optimize quietly in between. Realtime visibility changes that dynamic. Performance management through continuous goal tracking and lightweight check-ins makes progression and blockers visible as work happens.
This reduces the need for employees to “work around” formal processes just to stay on track. When outcomes are clear and shared, how the work gets done no longer needs to stay hidden.
Make feedback and progress cumulative
Unsanctioned behavior thrives in environments where feedback is episodic and disconnected from day-to-day work. Shifting to cumulative feedback captured through consistent documentation and shared reflections creates a clearer narrative of contribution over time. Employees are less likely to hide how they work when they can see how effort, judgment, and improvement compound. Transparency increases naturally when progress isn’t reset every quarter.
Train managers to guide AI-assisted workflows
Managers sit at the center of visibility, but many haven’t been equipped to lead AI-assisted work. Rather than asking them to enforce rules, train them to guide judgment. They should know when AI helps and where human review matters, so employees can adapt to the demands of their work without having to go behind IT’s back.
Performance conversations supported by check-ins and goal tracking give managers a practical way to engage their team and coach outcomes instead of monitoring tools, bringing work into the open without slowing it down.
Shadow AI is a symptom of distrust, not the cause
It’s easy to assume malicious intent when employees use shadow AI behind your back. More commonly, however, it’s a response to mounting work pressures and a lack of faith in your organization’s performance appraisal processes.
Banning shadow AI is most likely ineffective, but instilling a culture of transparency around workflows and performance management processes can eliminate the pressure to use unsanctioned tools in the first place.
If you want to learn more about the value of ongoing, transparent performance management for workplace culture, check out this piece on culture atrophy. If you’re more interested in exploring HR’s relationship with AI, this article on how automation forces HR to redefine its value is the one for you!