I’m attending Google Public Sector 2026 today, and some of the conversations have gotten me thinking about the benefits of AI and the risks that come with it.
Disclaimer: This is shared in a personal capacity, not to represent or critique any organization or product.
- People
Benefits: AI can catch things a tired person might miss, flag suspicious messages, and explain security issues to employees in plain language.
Risks: People are still involved in many security breaches. AI can help prevent mistakes, but it can also make mistakes and scams happen faster.
- Efficiency
Benefits: AI can help security teams sort through alerts, identify weak spots, and respond to potential threats faster.
Risks: Speed cuts both ways. When AI can act on sensitive data or systems, even a small error can have consequences before anyone has time to intervene.
- Customization
Benefits: AI tools can be tailored to an organization’s specific needs in ways that standard software may not allow. Done well, that can be a real advantage.
Risks: Custom-built tools can be harder to test thoroughly and review independently. The more a tool is capable of doing, the more important it becomes to understand what could go wrong.
- Connections
Benefits: Connecting systems can reduce manual handoffs, copy-and-paste errors, and delays. That can make everyday work much easier.
Risks: Connections also create potential points of exposure. Consider not just individual systems but how they interact.
- Countermeasures
Benefits: The more we understand how AI can be misled, the better we can prepare. Limiting access and requiring human approval for important actions are sensible safeguards for technology generally, not just AI.
Risks: An AI tool can be misled by a document or web page it reads, much as a person can be fooled by a convincing fake email. If that tool has access to real systems or sensitive information, the consequences can extend beyond the tool itself.
Potential Safeguards
- Limit access. Give each tool only the permissions it needs.
- Independent review. Have qualified third parties examine how systems are built and test how they behave.
- Review connections separately. Look carefully at integrations and the access they create.
- Establish approval points. Decide in advance which actions require a person to review and approve them.
- Plan for problems. Think through what happens when a system fails, is misled, or does something unexpected.
My takeaway is simple: AI offers enormous potential, and I’m interested in what it can help us accomplish. At the same time, it is worth thinking carefully about how we use it, what we allow it to access, and what safeguards we put in place.
Overall, though, I feel happy about the future. I think it’s bright.
