I love using AI. It's become a valuable part of my daily workflow, helping me explore ideas, research questions and work through challenges. At Radiant and Bark, we have policies that guide how we use it, including how we handle confidential information.
But one thing deeply concerns me: how quickly people are giving AI platforms access to their email and other sensitive parts of their lives.
I understand the appeal. An assistant that sorts messages, summarizes conversations and drafts replies can save considerable time. Before we connect everything, though, I believe we need to consider whose information we’re handing over.
Your inbox contains more than your own words. It may hold confidential client discussions, employee concerns, financial documents, family matters and private conversations with your loved ones. Someone shared that information with you because they trusted you.
Jesus commands us to love our neighbours as ourselves (Mark 12:31). That command should shape how we handle their private information, too. Before giving an AI platform access to our inbox, have you considered what giving another platform access means for them?
There are questions worth answering before clicking “Allow.” What can the tool access? What does it retain? Where is that information processed and stored? Who else can access it? What happens when you disconnect?
My personal feeling is that these answers deserve more attention than a general assurance that a “service is secure.”
We've already seen what happens when sensitive information is inadequately protected. Back in 2023, the breach at 23andMe affected almost seven million customers. Canadian and UK regulators found that information stolen in the attack included sensitive personal details, and some of it was offered for sale online. Although 23andMe is not an AI platform, its breach is a sobering reminder of the consequences of entrusting deeply personal information to an online service.
AI services have had their own failures. Last January, security researchers at Wiz discovered an exposed DeepSeek database containing more than a million log entries, including chat histories and API keys. DeepSeek secured the exposure after being notified. This was an exposed database, rather than evidence of a confirmed criminal theft, but sensitive information was accessible without authentication.
These incidents don’t establish that every platform will be breached, but they do give us reason to take that possibility seriously.
Imagine having to tell a congregant, donor, friend or family member that a private conversation had been exposed through a service you chose to connect to your inbox. Imagine that information being circulated or sold on the dark web. “I was trying to save time” would be a difficult explanation to offer.
That’s why I believe we need personal firewalls: clear boundaries around what we will allow AI to access. For me, that means being able to say, “I’ll use AI to explore an idea, help me with a draft or work through a problem. But I won’t give it unrestricted access to sensitive correspondence simply because the feature is available.”
For a specific task, it may mean removing identifying details or providing only the information needed. For an integration, it means evaluating the permissions, protections and suitability before connecting it. Sometimes, it means deciding that certain information stays outside the tool.
I believe we should use AI thoughtfully and confidently. I also believe our enthusiasm needs to be accompanied by sound judgment.
Before giving an AI platform the keys to your inbox, think about the people whose information is inside it. Their trust should help determine where you draw the line.
--
Helping organizations work through decisions like these is part of what we do at Radiant. Technology choices deserve strategic consideration before they create problems for your organization or the people who trust you. Through Radiant ONE, you can add our team to yours, bringing experienced marketing, communications and technology leadership into those conversations.
