They Just Connected It to Everything: The Problem with AI

First Contact is a managed IT service provider (MSP) based in Whitefield, Manchester, founded in 2001, offering IT support, cybersecurity, and AI governance advisory for growing businesses.

Nobody had snuck anything in. They'd just been sold the pitch that this will revolutionise how you work.

What Happened?

A client of ours got in touch recently about a technical problem with Outlook. A fairly routine callout - the sort of thing that's normally a five-second fix. As it happens, they were in good company: Artemis II Commander Reid Wiseman had the exact same complaint roughly 90,000 miles from Earth, and Mission Control's solution was the same one we reached for — remote in and sort it out.

Unlike NASA, though, remoting in for us didn't end with "Outlook's back online." What we actually found, once we got into it, was rather less routine: an AI tool had been connected to the business's CRM - with sweeping access to sensitive business and customer data — with no real oversight, no formal governance, and nobody fully aware of just how far that access extended.

The Problem with AI

Believing the hype is not the same as understanding the risk

Automating the CRM with AI is a genuinely sensible idea. The trouble is, "sensible idea" is not the same as "informed decision." Nobody sat them down and explained that AI this deeply integrated needs the same scrutiny as any other system handling sensitive data. So nobody asked.

That gap opened up a fair few concerns:

• AI holding sweeping access to sensitive business and customer data inside the CRM

• Very little visibility into how that data was actually being processed

• Commercially sensitive information potentially exposed to third parties

• Real questions around customer data protection and GDPR compliance

• No formal AI governance policy anywhere in sight

The business had assumed they were safe because the AI had arrived bundled inside existing software, rather than installed directly. Understandable logic, born of genuine enthusiasm rather than carelessness. Also, unfortunately, still a security breach.

There was, however, one thing that didn't go wrong. The business had also tried extending the same AI tool into their email, to automate triage and flag which messages needed a response first. That attempt hit a wall — because security we'd already put in place, precisely to limit this kind of exposure, stopped the connection going through. One dodged bullet, entirely by design rather than luck.

What we did

Rather than firefighting a single technical glitch, we treated this as what it actually was: a governance problem wearing a tech-support costume.

• Assessed the security implications of the AI's existing connection to the CRM

• Reviewed exactly what permissions and data access had been granted, and to what

• Confirmed why the attempted email connection had been blocked, and what that told us about the wider risk

• Flagged the compliance risks before they became compliance incidents

• Sat down with decision-makers and talked through what responsible AI adoption actually looks like — no scare tactics, just straight talk

• Mapped out a clearer, more secure path forward for using AI in the business going forward

  
  

What Changed?

The client walked away with a genuinely clear-eyed understanding of the risk they'd unknowingly been carrying, and a live example of what good security architecture actually buys you, since it had already quietly stopped the situation getting worse.

We also rolled out new cybersecurity tooling to proactively monitor AI usage across the business going forward, so any inadvertent leak of business data gets caught in real time rather than discovered months later during an unrelated callout, and made this monitoring available for all of our clients.

Potential compliance headaches get caught before they turned into actual, expensive ones. Data governance stopped being an abstract concept and became something the business could actually explain, and defend, if asked.

Key Takeaway

AI absolutely delivers business value, nobody's disputing that. But "this will revolutionise how you work" is a sales pitch, not a security assessment, and "it came bundled with software we already use" is not the same thing as "it's been properly assessed." Believing the hype isn't a moral failure. But it is, unfortunately, still a breach, just one built from enthusiasm instead of malice. The right security setup won't stop every mistake, but it will stop some of them before they happen, which is exactly what it did here. Convenience and security are different conversations, and the businesses that thrive are the ones who have both of them, ideally before something goes wrong rather than after.

FAQs

Can a security breach happen without anyone doing anything malicious?

Yes. Most AI-related data exposure isn't caused by hackers, it's caused by legitimate tools being granted more access than anyone reviewed, often because a business trusted a vendor's promises without asking what that access actually involved.

Is connecting AI to a CRM a GDPR issue?

It can be. If AI tools connected to your CRM are processing personal or customer data without you knowing what's collected, where it goes, or who else can access it, you may not be meeting UK GDPR accountability and data-processing requirements.

Can good IT security stop an AI tool from connecting to sensitive systems?

Yes, and it should. Properly configured access controls can block an AI tool from reaching systems like email or finance platforms even if someone in the business tries to connect it, giving you a safety net for exactly this kind of well-meaning mistake.

How do I create an AI governance policy for my business?

Start with an audit of every AI tool currently touching business or customer data — then define what access is acceptable, who approves new AI integrations, and how compliance is monitored going forward. An MSP can run this assessment for you if you don't have the resource in-house.

Can you monitor AI tools for data leaks after they're already connected?

Yes. Dedicated cybersecurity tooling can watch how AI tools are using business data on an ongoing basis, flagging unusual or risky activity as it happens rather than relying on someone stumbling across the problem later.