In the LoopIssue 5, 1 Jul 2026All issues →

Shadow AI is the security hole your board isn't talking about

Unapproved AI tools sit outside your security systems. Bans only drive them deeper underground, so the fix starts with an approved tool people want to use.

Here's a prediction I'd rather be wrong about. Shadow AI will fuel a 25% rise in enterprise cyber attacks by 2027. Unauthorized AI tool use has surged over the last two years (by 156%), and it's clear that most businesses still haven't clocked the worrying implications for their security.

How failed pilots drove AI underground

Let me explain how we got here. Companies rushed out AI pilots, employee adoption struggled, and leaders wrote the projects off as failed experiments. But the demand for AI stayed and your people didn't stop using AI. They just stopped telling you about it.

That's what Shadow AI is. Your sales manager pasting a client contract into a free chatbot. The analyst uploading sensitive customer data to a browser plugin nobody in IT has heard of, let alone approved. Each instance feels harmless in the moment, but is a door left open.

And hackers are about to walk right in. The problem with unauthorized tools is that they sit outside your security systems and governance processes, making them a perfect entry point. Ransomware attacks climbed 21% in 2025 alone, and soaring shadow AI use is about to push that figure higher in 2026.

Give employees an approved AI tool they trust, train them until adoption sticks and remove the incentive to operate in the shadows.
Give employees an approved AI tool they trust, train them until adoption sticks and remove the incentive to operate in the shadows.

Why banning the tools backfires

So what can businesses do once they face up to the threat?

The usual response is to ban the tools. I think that's the worst available option. Bans don't reduce shadow AI use, they just prompt employees to hide it better. And it's very difficult for businesses to defend an attack surface they can't see.

The cheapest security investment you'll make this year is giving your people an AI tool they don't have to hide.
John Margerison

That's why I believe the solution starts with adoption, not restriction. Rushed AI rollouts are what push workers toward unauthorized tools in the first place. To prevent this, businesses need to make their rollouts targeted and measured and train people properly until the authorized tool is the easy choice. Keeping employees on approved tech is the single most important factor in all of this.

Shoring up the defences that remain

Yet this still won't be fail-proof, so companies should also shore up their defenses. Audit every AI tool in use across the business regularly. Require workers to anonymize any data they put into AI tools, so sensitive information never touches an external system in the first place. This is exactly the reason we built redaction into the spine of our own product at XFactorAi, rather than treating it as an afterthought.

I'd add one more thing that rarely makes the security checklist. Ask your employees what they're actually using and why. Shadow AI is a symptom, not a disease, and the underlying condition is that your official tools aren't doing what people need them to. Fix that and the shadow shrinks on its own.

The action list for leaders

So here's the action list. Audit what's in use now. Roll out AI in a measured way and train people until adoption sticks. Anonymize data at the point of entry. Keep humans reviewing the important decisions. Get those basics right, and you make yourself a much harder target. Get them wrong, and shadow AI quickly becomes an open door for cyber attackers.

Trust me when I say that the cheapest security investment you'll make this year is giving your people an AI tool they don't have to hide.

Keep reading. It's free

Still to come in this issue

  • Why banning the tools backfires
  • Shoring up the defences that remain
  • The action list for leaders

2 more minutes to read.

Read by 1,000+ founders & CEOs worldwide. No spam, ever. One click to unsubscribe.

The reading list

Three things John has been reading, and why they're worth your time.

  1. itweb.co.zaAI breaks through to the other side

    My thoughts on rogue AI – in the wake of an OpenAI model breaching a test environment – were included in this article for ITWeb.

  2. BBCOpenAI says its AI went rogue and launched 'unprecedented' cyber-attack

    OpenAI's own model escaped its testing environment and launched an "unprecedented" cyber attack. This should be a warning to enterprise leaders rushing to hand increasingly autonomous tools access to their systems.

  3. AxiosGoogle DeepMind's Demis Hassabis calls for U.S.-led global AI watchdog

    Demis Hassabis urges the U.S. to create an AI watchdog capable of testing the world's most advanced models and coordinating an industry-wide slowdown if risks escalate.

Questions this issue answers

What is shadow AI and why is it a security risk?

AI management expert John Margerison describes shadow AI as employees using AI tools their employer hasn't approved, such as pasting client contracts into free chatbots or uploading customer data to unvetted browser plugins. Because these tools sit outside security systems and governance processes, he argues they become a perfect entry point for attackers.

Should companies ban unapproved AI tools?

AI management expert John Margerison calls banning the tools the worst available option. He says bans don't reduce shadow AI use, they just prompt employees to hide it better, leaving businesses unable to defend an attack surface they can't see.

How can businesses reduce the risk of shadow AI?

AI management expert John Margerison recommends auditing the AI tools in use, rolling out AI in a measured way and training people until adoption sticks. He also advises anonymizing data at the point of entry, keeping humans reviewing important decisions, and asking employees what they actually use and why.

Written by

John Margerison

Founder and CEO, YesBusiness AI

Roughly monthly. Unsubscribe any time.

Get in touch

For press, speaking or working with John.