We recently reran our AI Proficiency Survey (read the full report here) and found that 11% of companies still explicitly ban the use of AI and another 26% are silent on AI, with no explicit point of view or policy. Yet nearly half of employees in companies with AI bans still use AI regularly, and two-thirds of employees at companies that are silent do so.

Here’s what shadow AI looks like, and which industries have the most secret AI users.

Shadow AI means less safe and effective AI usage

43% of knowledge workers in companies that ban or prohibit AI use it regularly. Nearly 6% use it daily, and another 17% use it at least once a week. In companies that are silent on AI, 64% of employees are using it, with at least 33% using AI weekly.

Of these AI users, only 23% use a paid tool. That means that the majority of shadow AI is done with limited free models that provide less nuanced answers and less privacy and security controls.

Shadow AI users are also less effective at using AI than their counterparts in companies that explicitly promote and encourage AI use. On the one hand, they feel less proficient – only 10% self-report as advanced AI users, compared with 16% of employees in companies that approve of AI.

On the other hand, they ARE less proficient based on our methodology. Just 4% of employees using AI at companies that ban or are silent on AI are AI Experts or AI Practitioners (the top 2 most proficient parts of the workforce). Among companies that approve of AI, 15% of employees using AI are AI Experts or Practitioners.

Shadow AI users know less about how to use AI safely and are worse prompters. The average Shadow AI user is in the 44th percentile for AI knowledge compared to the 52nd percentile for an AI user at a company that approves of AI. And the average Shadow AI user is in the 26th percentile for AI prompting skills vs. the 34th percentile for the average AI user at a company approving of AI.

So it's the worst case scenario – if you ban or are silent on AI, you can’t stop employees from using it. But they will use it less safely and less effectively than if you explicitly approve of AI use – likely because approval is correlated with rolling out a paid LLM and AI training for employees.

Industries with the most shadow AI

While our survey shows that there’s shadow AI in many industries, these are the ones it’s most prevalent in:

Two themes emerge

1. Industries like healthcare and finance that deal with highly sensitive information.  The security risk from using AI is heightened in these industries – hence the bans. But we’ve just proven this doesn’t stop usage, it just makes these businesses more susceptible to security risks. Plus, companies like Moderna have proven that there are significant benefits to AI in these industries, opening the door for experimentation.
   
2. Industries like retail and food & beverage, with a lower percentage of desk workers. In these industries, language-intensive functions like marketing and engineering are likely where the most shadow AI is happening.

What to do about shadow AI

AI use is happening in your company whether you want it to or not. The real problem is two-fold:

1. These users aren’t proficient in safe or effective AI use. This introduces more risk to the organization.
   
2. They’re using free models. Not only are they not getting the best insights from AI, they’re using tools that don’t allow them to customize how their data is used – which, again, is a risk for your company.

The most important thing for leaders to do now is create policies and guidelines around how AI can be used and what data can be shared with it. Then provide training on how to use these tools safely and what to avoid.

Banning it doesn’t work, so it’s time to make sure your team is as proficient as your competitors. Download the full AI Proficiency Report to uncover the 3 other company policies that impact AI usage and skill.