The AI Use Cases That Look Harmless May Carry the Most Risk

An AI system that affects human safety usually gets attention. People see the stakes, bring in the right reviewers and ask what could go wrong.

A résumé-screening feature often gets none of that. Someone in HR has hundreds of applications to get through, sees an option that will speed things up, and switches it on. They aren’t trying to take a big risk. They’re trying to make a routine task manageable. But the feature may run on a sophisticated AI model, and that raises questions about discrimination, privacy, consent and disclosure.

Dan Clarke, President of Truyo and IntraEdge Solutions, says the gap between how risky something looks and how risky it is causes a lot of companies trouble. His team surveyed public-sector organizations and found that respondents generally rated systems tied to human safety as high risk. Résumé screening and chatbots were often rated as almost no risk at all.

That rating decides how a use case gets handled. When people see risk, they look into it. When a feature feels ordinary, it can go live without anyone asking what data it uses, whether the organization has permission to use that data, who could be affected, or what happens when the system gets something wrong.

Clarke joined Sachin Kamdar and Doyle Irvin on Building For Others to talk about why these quiet, everyday deployments need more attention, and how enterprises can govern AI and still keep its competitive benefits.

How Truyo Helps Companies See and Govern Their AI Use

Truyo’s approach to AI governance came out of its work on enterprise privacy and consent. Intel originally recruited Clarke to help companies get ready for GDPR, and he sees a lot of overlap between the privacy questions of that time and the governance questions companies face now.

It starts with data. Before you judge what a model can do, you need to know whether you had permission to use the data behind it. You also need to know where AI is running in the organization. That’s hard when AI features come built into software employees already use and can be turned on without any formal project.

Truyo’s platform helps companies find their AI use cases and keep an inventory of them. Once a company knows it’s using AI for résumé screening, customer support, image recognition or some other workflow, it can assess the business and regulatory risk. It can record how the decision was made, trace consent and data permissions, put controls in place, and give the people affected the notice they’re owed.

Clarke treats this inventory as the foundation. Many organizations hand AI oversight to IT, but a lot of the important use cases show up in HR, marketing and finance. People in those departments may not know a feature counts as AI, or that turning it on brings new obligations. Truyo pairs automated scanning with the company’s existing privacy infrastructure so more of that activity becomes visible.

The platform uses fixed rules and AI-based analysis side by side. Rules work well for structured data like Social Security numbers or credit card numbers. They work less well when sensitive information shows up in everyday language. A sentence about not being able to ride an exercise bike because of a broken leg can contain health information, even though it matches no tidy numeric pattern. Semantic analysis catches context that traditional compliance rules miss.

Gartner recently named Truyo a Leader in its inaugural Magic Quadrant for AI Governance Platforms. Of the companies evaluated, Truyo sits in the Leaders quadrant with IBM and ServiceNow. Clarke credits part of that to how Truyo treats privacy, consent and AI governance as one program instead of three.

The Rest of the Discussion

  • Many AI lawsuits are really about data permission. Clarke argued that cases about chatbots, model training and automated screening often come down to old privacy and consent questions. The AI may be new, but the real issue is usually whether the company had the right to use the data that way.
  • The United States needs one federal standard. A patchwork of state laws may drive demand for automated compliance platforms, but Clarke still backs federal legislation. Companies trying to comply in good faith face 21 state laws, or 22 for social media companies, and the laws differ in ways that matter.
  • AI enforcement may be tougher than privacy enforcement. Clarke described a conversation with a state attorney general who said voters care about AI in a way they never cared about privacy. That puts pressure on elected officials and regulators to act. Clarke expects a big wave of US enforcement, and thinks Europe may enforce the EU AI Act as hard as GDPR, or harder.
  • Open-weight models bring more options and more governance work. They widen access and give companies choices, but changing a model’s parameters can make bias, hallucination and guardrail problems harder to control. Clarke focuses on governing how a model is used, especially where employment, health, housing, finance or human safety are involved.
  • Even a model’s creators may not know how it gets its answers. Clarke told the story of a facial-recognition system that, without anyone planning it, compared features of human faces with traits of different dog breeds. It was fast and accurate, and its creators couldn’t explain why it had landed on that method. Governance can’t assume model behavior will always be explainable.
  • It may take AI to govern AI. Fixed rules still matter and should stay switched on. But no rule can anticipate every way a person might phrase sensitive information. Clarke thinks AI-based sentiment and semantic analysis can fill that gap, especially as enterprise AI use gets too varied and fast-moving for people to review by hand.
  • Disclosure is one of the easiest protections you have. Companies sometimes avoid saying that AI is screening résumés or running a chatbot. Clarke says do the opposite. Telling applicants and customers clearly gives them useful information and puts the company on firmer legal ground than hiding AI’s role.

Clarke isn’t telling companies to slow down on AI. He thinks it’s a competitive advantage they need to take. His warning is about timing: if you leave governance until after something goes wrong, you risk lawsuits, regulatory action and a loss of confidence inside the company that can stall the whole program.

The safer route starts with visibility. Know where AI is in use, understand the data and the people involved, write down the risks, and put controls in place before a feature that looked harmless turns into a serious problem.

About the guest: Dan Clarke is President of Truyo and IntraEdge Solutions and a nine-time CEO. His work covers enterprise privacy, consent and AI governance. Learn more about Truyo and connect with Dan on LinkedIn: https://www.linkedin.com/in/danclarke/

Bring your agent. Bring your team.