How to use generative AI to improve operations while protecting sensitive data
Connecting state and local government leaders
COMMENTARY | Large language models like ChatGPT offer state and local agencies exciting new capabilities, but they also introduce new security risks.
ChatGPT captured the public’s imagination with neat tricks like churning out grocery lists in the style of a Shakespeare sonnet. But the generative artificial intelligence tool has the potential to revolutionize business processes. It can write reports, provide customer support, perform data analysis, summarize market research, develop training materials, automate emails, schedule appointments and much more. It’s already being deployed across industries as diverse as technology, finance, e-commerce, media, law and health care.
Public and private sector agencies recognize that generative AI is a powerful tool that will be affordable, ubiquitous and transformational—but not without risk.
Today, generative AI produces outputs primarily based on the vast public datasets commercial solutions were trained on. But agencies that want to use the technology to analyze budgets, boost customer experience with personalized information, or query in-house knowledge bases will have to train their generative AI solutions using their own proprietary information. That raises a host of cybersecurity concerns.
AI outputs based on internal data could increase the risk of leaking sensitive information to malicious actors. They could also expose employees’ and citizens’ personally identifiable information, placing their privacy at risk.
Here are three effective strategies agencies can use to take advantage of generative AI while protecting their sensitive data:
1. Implement a comprehensive data classification strategy. If organizations intend to leverage their data stores as generative AI inputs, then it’s incumbent on them to precisely identify which data is safe to include and which is so sensitive that it must be excluded. Achieving that goal begins with systematic data classification.
The federal government has been classifying data for a long time, and its experience offers useful guidance to state and local agencies on how they can classify their data. Of course, federal civilian and military agencies have their Secret and Top Secret classifications. But in general, organizations should categorize and tag data based on its value to the organization, its criticality for internal operations or for serving the public and its sensitivity should that data become exposed.
2. Set policies and educate employees on “shadow AI.” When employees or teams in an agency use an AI system that isn’t formally approved or managed by the organization, they’re using what’s called shadow AI. Because these AI systems lack managerial oversight, integration with other systems and compliance with policies, they can expose an agency to security, operational and regulatory risks.
To protect against shadow AI, set clear policies for the use of AI and communicate those policies to all employees. Establish a process for teams to request AI capabilities from the IT department and receive formal approval. Make approved AI tools and training available so that teams don’t feel the need to go around AI policy. And monitor network and data access to detect unauthorized AI use.
3. Practice good generative AI governance. Create a formal policy on approved and prohibited generative AI use cases, as well as on the internal data that may be used in AI applications or to train AI models. A generative AI governance board with representatives from the IT, HR and legal departments can help agencies formally assess use cases and associated data.
If there’s one constant in emerging technologies such as generative AI, it’s that they’re always changing. The same could be said for cyberthreats.
The power of data-centric security is that as capabilities and risks evolve, data protections persist. Agencies can benefit from generative AI while protecting privacy and preventing sensitive data from falling into the wrong hands, thereby keeping agency, employee and citizen data safe.
NEXT STORY: Funding for local cybersecurity efforts is insufficient, survey says