Third-Party Apps and AI in Microsoft 365: The Risks Nobody Reads About at Sign-Up

Who's Actually Connected to Your Microsoft 365 Tenant? (Besides You, Obviously)

Let's play a game. Think of every third-party app and AI tool your team has connected to Microsoft 365 over the years.

Now think of how many of those you personally vetted.

If your answer is "there's a Dave in accounts who set up an AI email thing in 2024 and he's since left," or "it was me, at 11pm, in a hurry, and I used the app twice" then this blog is for you. Hello. Welcome. Get comfy with a cuppa and a biscuit.

It Started With Such Good Intentions

Nobody wakes up wanting to compromise their data security. But someone signs up for a clever little AI tool that promises to summarise emails, the sign-up screen flashes "Allow access to Mail, Files, Contacts and Calendar?", and it's clicked before the kettle has boiled.

And that's the trap. That innocent-looking permissions screen isn't asking politely. It's handing a set of keys to the entire filing cabinet, including the dusty drawers nobody has opened since 2019. Historic data included, whether or not anyone remembered it was there.

The tool itself might be perfectly lovely. The company behind it might be scrupulous. But you have just let a company you've never met wander through your business records, and under UK GDPR, you're the one who signs for the consequences if they drop something.

Where AI Sticks Its Nose

Right, the AI bit. And here we're not trying to frighten anyone, just be honest about the plumbing.

A regular app stores your data. An AI service reads it. Properly reads it. It will happily:

  • Summarise the conversation threads you'd forgotten about

  • Notice that Carol in HR messages Finance every Tuesday about the same sensitive matter

  • Draw conclusions from patterns across everything it can see

None of that is malicious. It's just... thorough. A bit like inviting a charming stranger to a party, only to find out they've memorised the guest list, the seating plan, and who isn't speaking to whom.

Two practical questions worth asking any AI provider, in writing, before you connect:

  1. Do you use our data to train your models? Because data absorbed into a model can't realistically be deleted afterwards. What's learned can't be unlearned.

  2. Where is our data stored and processed? Sometimes outside the UK, sometimes somewhere with rather different data protection habits.

Most reputable providers have good answers to both. But "most" is doing some heavy lifting in that sentence, so do get it in writing.

Other Bits Worth Knowing (The Fine Print, Translated)

  • Disconnecting isn't deleting. Removing an app stops future access; it doesn't hoover back everything already shared. Retrieval often needs a formal request, and sometimes patience.

  • Terms can change overnight. Providers can update their terms, pricing and data practices whenever suits them. Your annual review isn't paranoia; it's hygiene.

  • Sub-processors exist. Your friendly app supplier may have suppliers of their own, like Russian dolls, except every doll is another company with its hands on your data.

  • Privileged material deserves a VIP door policy. Legally privileged or highly confidential content going through external AI tools can compromise its protected status. Best kept well clear.

  • Every connected app is another front door. More doors, more chances one gets left unlocked. Not a reason to barricade yourself in, but a reason to know where your doors are.

What "Good" Looks Like‍

The reassuring news: none of this means avoiding helpful tools. It means connecting them with your eyes open. Our recommended routine:

  1. Keep an inventory. A simple list of every connected app, ideally in your Article 30 GDPR records. Yes, this means finally writing down what Dave installed, and what you installed yourself and forgot about.

  2. Ask before granting, not after. Where does data live? Is it used for anything else? What happens to it if we cancel?

  3. Give the minimum permissions. If a scheduling tool asks to read your email archive, it's not being thorough, it's being greedy.

  4. Do a Data Protection Impact Assessment (DPIA) for AI tools touching personal data. It's shorter than you fear and worth far more than it costs.

  5. Review regularly. Apps that were essential in 2023 might be dead weight now. Retire them with dignity.

  6. Make it safe for staff to ask. A culture where someone flags "should I connect this?" beats a culture where they quietly do it and hope.

To keep things grounded, we ask all our clients to sign a Third-Party App and AI Integration Risk Acknowledgement Form before connecting anything new. If you are a client of ours with Microsoft 365 services through us and you have not received one yet, it is winging it’s way to you!

The Bottom Line

Third-party and AI tools are genuinely useful, and pretending otherwise helps nobody. The goal isn't a locked-down tenant where nothing works. It's a tenant where everything connected was chosen deliberately, checked properly, and remembered fondly.

So finish that cuppa, and give Dave's legacy integrations, and your own, a kind but firm retirement. If you'd like to review what's connected to your tenant, drop us a line and we can do a quick audit with you, and help clear out anything that is no longer useful!

 
Ellen Badat

As a web designer based in Gloucester, I have been designing websites for over 20 years by taking my clients ideas and turning them into great looking websites.

https://purpletree.solutions
Next
Next

Passkeys: The End of Passwords (and the Traps Nobody Warned You About)